5G—Next-Generation Wireless KEY POINTS

EQUITY RESEARCH

INDUSTRY UPDATE

Oppenheimer & Co Inc. 85 Broad Street, New York, NY 10004 Tel: 800-221-5588 Fax: 212-667-8229

Rick [email protected]

Wei [email protected]

Disseminated: October 2, 2019 06:01 EDT; Produced: October 2, 201906:01 EDT

For analyst certification and important disclosures, see the DisclosureAppendix.

October 2, 2019

TECHNOLOGY/SEMICONDUCTORS & COMPONENTS

5G—Next-Generation Wireless5G Technology PrimerSUMMARYThe continued adoption of connected devices and exponential growth of mobile datatraffic have put unprecedented strain on current wireless infrastructure networks.In a race to deliver greater performance, monetize the thirst for data, and slowlytransition from a commoditizing, consumer-driven subscriber base to a high-value,mission-critical, enterprise-driven subscriber base, wireless service providers havebegun to accelerate the pace of 5G New Radio (NR) network deployments acrossthe globe. For this report, we interviewed more than 25 companies across thewireless infrastructure and device supply chains and conclude that the 5G upgradecycle will drive both a cyclical rebound in carrier capex and structurally highersemiconductor content per macro cell, which will drive significant growth for manyof the infrastructure-related names in our coverage universe. The adoption of 5Ghandsets is likely to shorten replacement rates, stabilize handset units, and drivesignificant growth in the radio frequency (RF) TAM.

KEY POINTS

■ Carrier capex on the global mobile RAN market bounced between $30B and $35Bfrom 2004-2016 but has hovered closer to $28B. The drop is largely attributableto carriers slowing spending post aggressive 4G build-outs—particularly in China(45% of global 4G BTS), where spending fell nearly 50% peak-to-trough. Thatsaid, we expect a cyclical recovery as 5G roll-outs begin globally and view $33B-$34B, or a 20% bounce from current levels, as a likely scenario.

■ 5G networks promise a 1000x increase in capacity via a 10x increase inthroughput and support for 100-fold increased connections per unit area whilereducing cost per GB by as much as 35x. To achieve such a high bar, thecomplexity—and thus, semiconductor content within macro cells—is going toincrease significantly, esp. with M-MIMO implementations. We expect 2-3x digital,2-4x analog, and 3-5x RF content across the baseband (BBU) and radio (RU) unit.

■ While we’ve seen dramatic changes across the wireless network equipmentmanufacturer (NEM) landscape over the last 15 years with Huawei/ZTE (both relyon internal silicon arms) gaining ~40pts of share, we see potential for reversalduring the 5G era, esp. as US sanctions impede Huawei’s early progress. The USand Japan elected not to use Chinese NEMs, South Korea has reduced relianceon Huawei, and Samsung is becoming more competitive across the globe.

■ Smartphones is a $483B market in 2018 with 1.4B units shipped. Units have beenflat-to-LSD(%) decline the last two years as we near the tail end of the 4G cycle.5G is a growth catalyst led by RF, and we estimate TAM of $26B by 2025 (8%CAGR from 2018). Due to rising band counts/complexity in 5G, we see filters asthe greatest content opportunity in RF with BAW filters growing 13% CAGR to$8.9B by 2025.

■ As technology transitions from 4G to 5G, companies in the supply chain arerealigning strategies. In baseband/AP, technology gap between tier-1 handsetOEMs and merchant component suppliers MediaTek/Qualcomm is closing whendesigning SoC solutions. We see more tier-1 OEMs sourcing back-end 5G-components in-house. As a result, QCOM is differentiating by developing a fullmodem-to-antenna solution. However, we expect limited success as handsetOEMS have expressed interest in working with longtime higher performance RFincumbents AVGO, QRVO, SWKS.

Table of Contents 5G INTRODUCTION .............................................................................................. 3

5G AT A GLANCE – WHAT IS 5G AND WHY DO WE NEED IT? ...................................... 3 NEW FREQUENCIES WITH MORE BANDWIDTH .......................................................... 5 MASSIVE MIMO AND BEAM FORMING ..................................................................... 8 DELIVERING IMPROVED CUSTOMER EXPERIENCE ................................................. 10 NOT ONLY ABOUT THE CONSUMER ...................................................................... 11 CARRIER CAPEX CYCLE ...................................................................................... 12 BASICS OF A MACRO CELL—BBU, RRH, AND ANTENNA ...................................... 15 MACRO CELL TAM: BB & RRH UNIT AND ASP ANALYSIS .................................... 19 MMWAVE: SMALL CELLS RAMP-AGAIN? ............................................................... 21

TELECOM EQUIPMENT ..................................................................................... 23 ERICSSON ......................................................................................................... 23 NOKIA ................................................................................................................ 23 SAMSUNG ......................................................................................................... 24 HUAWEI ............................................................................................................ 24 ZTE .................................................................................................................. 25

INFRASTRUCTURE OEMS ................................................................................ 25 ANALOG DEVICES .............................................................................................. 25 MARVELL ........................................................................................................... 26 QORVO .............................................................................................................. 28 TEXAS INSTRUMENTS ......................................................................................... 29 INTEL ................................................................................................................. 30 NXP .................................................................................................................. 31 MAXIM ............................................................................................................... 32 XILINX ................................................................................................................ 32 INFINEON ........................................................................................................... 33 SEMTECH ........................................................................................................... 33 LATTICE ............................................................................................................. 33 INPHI ................................................................................................................. 34

SMARTPHONES ................................................................................................. 34 SEMICONDUCTORS ............................................................................................. 34 5G SMARTPHONE BOM/TEARDOWN ................................................................... 36 APPLICATION PROCESSOR .................................................................................. 37 BASEBAND PROCESSOR ..................................................................................... 38

APPLICATION/BASEBAND VENDORS ............................................................ 39 APPLE................................................................................................................ 39 QUALCOMM ........................................................................................................ 39 MEDIATEK ......................................................................................................... 40 SAMSUNG .......................................................................................................... 40 HISILICON/HUAWEI ............................................................................................. 40 INTEL ................................................................................................................. 40

RADIO FREQUENCY FRONT END .................................................................... 41 FILTERS ............................................................................................................. 41 5G FREQUENCIES .............................................................................................. 42 FILTER TYPES – SAW, TC-SAW, IHP-SAW, BAW, FBAR ................................. 43 POWER AMPLIFIER, LOW NOISE AMPLIFIER, SWITCHES, TUNER, DIPLEXER ........... 45

RF FRONT-END VENDORS ............................................................................... 46 BROADCOM ........................................................................................................ 47 SKYWORKS ........................................................................................................ 47 QORVO .............................................................................................................. 47 QUALCOMM ........................................................................................................ 48 MURATA ............................................................................................................ 48 AKOUSTIS .......................................................................................................... 48 RESONANT ......................................................................................................... 48 KNOWLES .......................................................................................................... 44

ACRONYMS ........................................................................................................ 50

2

TECHNOLOGY / SEMICONDUCTORS & COMPONENTS

5G Introduction

5G at a Glance—What Is 5G and Why Do We

Need It?

The continued growth of both connected devices and traffic per device has caused the

amount of internet protocol (IP) traffic to continue to grow at an exponential pace—Global

IP Traffic in terms of Exabyte (EB) per month is growing >25% every year and essentially

doubling every three years. To put this in context, the amount of IP traffic generated from

2018-2024 will be ~4x the amount of IP traffic created in the history of the world prior to

2018. As such, data demands have strained and will continue to strain our current

wireless network infrastructure. While this is a problem, it’s not a new problem; in fact,

improved user experience, new applications, and network strain have been three of the

primary drivers behind the prior wireless network infrastructure transitions and

transformations for decades. As such, the Third Generation Partnership Project (3GPP)

began initial 5G New Radio (NR) studies in 2015 and completed the first set of 5G NR

standards in December 2017. 5G NR is expected to deliver more efficient communication

protocols, new and wider spectrum, and new technologies, which in combination will

provide more capacity, more throughput, lower latency, lower energy consumption, and

significantly lower cost per GB of data. Specifically, 5G promises to bring a 1000x

increase in capacity, a 10x increase in throughput, a 10x decrease in latency and support

up to 100x the number of connections per unit area given a 3x increase in spectral

efficiency, and a 100x increase in network efficiency relative to initial 4G networks all

within the same 4G power budget. More impressively, 5G delivers these improvements

while delivering up to a 35x reduction in cost per GB.

Exhibit 1. Global IP Traffic; Cost per EB by Technology

Source: Cisco, Mobile Experts, Oppenheimer & Co. Estimates

It’s worth noting that each country (and most network operators) will likely roll 5G NR out

in two distinct phases—phase 1 will be 5G non-standalone (NSA) while phase 2 will be 5G

standalone (SA). Initial 5G NSA deployments and the first commercial launches occurred

in 2018, whereas the SA standards are not expected to be finalized until the end of this

year.

Note that the first operational 5G NSA network was demonstrated at the 2018 Winter

Olympics in Pyeongchang, South Korea. The initial demonstration showed a fixed wireless

access (FWA) use case where the 5G network streamed multiple 4K videos from up to 80

cameras to the Olympic audience simultaneously. Further, all three major carriers in South

Korea—SK Telecom (SKT), Korea Telecom (KT), and LG Uplus (LGU) —launched nation-

3


wide mobile 5G services in April 2019 with more than 2M aggregate subscribers to date

and current expectations for 4-5M by year-end.

The primary difference between NSA and SA is that SA requires a carrier to completely

over-haul its core, transport, and radio access networks (RAN). Conversely, NSA only

requires a RAN upgrade—Baseband Units (BBUs), Remote Radio Heads (RRHs), and

Antenna—so the new 5G RAN is simply paired with the existing 2G/3G/4G core and

transport networks. As such, networks will take a big step forward in terms of throughput

and capacity during the NSA phase as operators leverage the latest and greatest RAN

technologies, as well as the incremental spectrum that has been and will continue to be

allocated for 5G NR. Simply put, 5G NSA will do exactly what each transition from 2G to

3G to 4G has done, allowed consumers to do progressively more with their mobile

devices.

Specifically, the 5G NR standards have been set to facilitate the adoption of Enhanced

Mobile Broadband (eMBB), which promises to bring features like augmented (AR) and

virtual (VR) reality to smartphones in 4K/8K resolutions. It’s worth noting that it would be

extremely difficult for an operator to move straight from 4G to 5G SA as: (1) that operator

would likely lose subscribers to other operators that elected to do a phased roll-out and

can thus advertise 5G capabilities; and (2) a phased roll-out makes 5G much more

palatable from a capex standpoint as carriers can stagger the capex peaks of their

respective core and radio access networks.

In our view, commercial 5G SA deployments are unlikely before 2021, will largely use the

same RAN as 5G NSA, and for all intents and purposes will go unnoticed by the common

user. However, the core network will undergo dramatic changes. The new 5G SA core

network will be virtualized from end-to-end as network operators transition away from fixed

function hardware appliances in favor of commodity servers—the transition to software-

defined networking (SDN) and network function virtualization (NFV) began in the middle of

the 4G cycle and, in combination, should dominate core network architectures as 5G SA

rolls out. This transition will allow operators to execute on more distributed (and

abstracted) RAN architectures, mediate the network in real-time via software, and deploy

machine learning (ML)/artificial intelligence (AI) across the network. Perhaps most

importantly, end-to-end network virtualization will enable network slicing—essentially

separate the capacity into multiple functional models, thus permitting custom services and

service levels tailored to specific enterprise or verticals, which, in combination with higher

levels of edge compute, unlocks the other two most hyped use cases (in addition to

eMBB) associated with the development of 5G NR: Ultra-Reliable Low Latency

Communication (URLLC) and Massive Machine Type Communication (mMTC).

URLLC refers to using the network for mission-critical applications that require

uninterrupted and robust data exchange such as autonomous driving, traffic safety &

control, remote manufacturing, as well as other applications across the industrial, medical,

and government complex. Conversely, mMTC would be used to connect an extremely

large number of scalable, low power, and low cost devices across a vast area such as

sensors for smart metering, fleet management, and all sorts of tracking applications. The

reason we called out 5G SA—and specifically the capability to network slice—as

particularly important is that both networking equipment manufacturers (NEMs) and

carriers view these applications as a means of accelerating revenue growth by moving

their businesses beyond the commoditizing consumer market, where carriers are facing

incremental competition from cloud vendors, and into high-value enterprise verticals.

While it remains to be seen that carriers will be able to drive material business value

above and beyond the temporary premium pricing models for 5G-enabled eMBB, the long-

term dream is for 5G service providers to be able to provision specific virtual networks

and/or services and sign service-level-agreements (SLAs) with companies across

4


verticals. It’s also worth noting that, given the timelines associated with 5G SA roll-outs

and more specifically timelines associated with robust, dense, and broad 5G SA network

coverage, we aren’t likely to know if NEMs and carriers will be successful in realizing this

vision until the 2025-2030 timeframe. However, it’s worth noting that, even if these

incremental revenue streams don’t materialize, 5G SA could prove to be a boon to

profitability for carriers as end-to-end virtualization should improve network utilization and

traffic management and ultimately lower costs.

5G: New Frequencies with More Bandwidth Shannon Theory is the underlying science that explains why new frequencies, particularly

frequencies with more bandwidth, accelerate the data rate and increase network capacity.

Shannon Theory roughly states that a network’s data rate is equivalent to its number of

channels multiplied by the bandwidth of the channels and then again multiplied by log2

(1+SNR) where SNR equals the signal-to-noise ratio. Channels are roughly equivalent to

the number of antennas (hence why networks are progressively moving from 1x1 to 2x2 to

4x4/8x8 and now 32x32/64x64), bandwidth is how wide the spectrum is (n79 = 4.4GHz-

5.0GHz so 600MHz bandwidth), and SNR is how high the signal power is above the noise

floor.

Bandwidth: the difference between the upper and lower frequencies in a continuous band

Channel Bandwidth: the FCC may allocate the regional available bandwidth to broad license holders so their signals do not mutually interfere.

As such, the 3GPP divided the specification for the 5G air interface into two ranges,

frequency range one (FR1) and frequency range 2 (FR2). FR1 is commonly referred to

as Sub-6GHz as it spans from 450MHz to 6GHz, and thus overlaps with and extends

2G/3G/4G frequencies. FR1 bands are numbered 1-255 and provide reliable, cost-

effective mobile broadband coverage. The maximum channel bandwidth defined for FR1

is 100 MHz, due to the scarcity of continuous spectrum in this crowded frequency range;

however, that’s still a significant increase relative to 3G/4G, which used individual

channel bandwidths of 5-20MHz.

FR2 is often referred to as millimeter wave (mmWave) and spans from 24GHz to 52GHz

with bands numbered from 257-511. Relative to FR2, the minimum channel bandwidth is

50MHz and the maximum is 400MHz, and thus basically 10-20x as wide vs. 3G/4G. FR2

frequencies were not used in any of the prior generations of networks (or smartphones);

thus, they have significantly less clutter vs. FR1, enabling wider bandwidths and extremely

high throughput, which is particularly useful in densely populated areas. That said, not

everything about mmWave is great. Given that FR2 is defined by significantly higher

frequencies and shorter wave lengths, it propagates less effectively than lower

frequencies. It offers relatively short-reach as it has trouble: (1) penetrating walls or glass;

and (2) performing in harsh weather conditions, i.e., rain or snow.

5


Exhibit 2. 5G Frequencies vs. 2G/3G/4G Frequencies

Source: Mobile Experts, Oppenheimer & Co. Estimates

From a hardware perspective, the implementation of new, higher frequency bands with

wider bandwidths is driving a shift in the base materials used in the underlying radio

frequency (RF) technologies. From an infrastructure perspective, the power amplifier (PA)

space is likely to undergo the largest transition from a materials perspective (in addition to

significant TAM expansion). Historically, LDMOS silicon technology has been the

dominant technology in the base station (BTS) PA market as it had the ideal power and

frequency characteristics at a relatively low cost.

GaAs: Gallium Arsenide

GaN: Gallium Nitride

GaN-on-SiC: Gallium Nitride on

Silicon Carbide

SiGe: Silicon Germanium

LDMOS: Laterally Diffused Metal

Oxide Semiconductor

Power is important because if more power is applied to a given RF signal over a given

frequency, the signal can travel a further distance. The average power out of a PA can

be calculated by the following formula [POUT =VDD2 / 2 x RL] where RL is the optimal

resistance load and is essentially fixed at 50 Ω, implying that the power capability of a PA

is essentially dictated by the voltage applied to it. Thus, in order to increase PA power,

the voltage needs to be increased and certain materials handle high voltages.

Specifically, LDMOS and GaN-on-SiC operate at either 28V or 48V; whereas GaAs

operates between 5-7V and SiGe operates between 2-3V. However, LDMOS has a

much more robust supply chain and lower cost vs. GaN-on-SiC, so GaN-on-SiC

represented only ~15% of the infrastructure PA market in 2018, as compared to LDMOS

at ~75% of the market.

However, the one area where LDMOS is weak relative to GaN-on-Sic, GaAs, and SiGe is

frequency. LDMOS is primarily useful below 4GHz (historically, the peak was thought to

be 2.6GHz, but NXPI has been able to stretch the use case of LDMOS to nearly 4GHz),

whereas the other three materials can handle frequencies up to100GHz. A material’s

ability to handle higher frequencies is related to the electron mobility of that given material,

and in terms of electron mobility, LDMOS lags behind other materials. GaN-on-SiC and

SiGe have similar characteristics at ~2,000cm/V.s while GaAs has the highest electron

mobility at ~5,000cm/V.s.

6


Exhibit 3. Semiconductor Materials—Power vs. Frequency

Source: Oppenheimer & Co. Estimates

As such, we should see a slow shift in the primary materials used in different sub-sections

of the PA infrastructure market—especially as 5G M-MIMO and mmWave help create

more robust GaN-on-SiC and SiGe supply chains and thus lower their relative costs over

time.

Exhibit 4. RF PA Material by Application Over Time

Telecom

Infrastructure

Multi

Market

x

2015 2016 2017 2018 2019 Future

LDMOS GaAs GaN

Macro Cells

Military

RF Energy

Small Cells

Backhaul

Wired/CATV

SATCOM

Radars

Source: Yole, Oppenheimer & Co. Estimates

In overview, we see explosive growth in the infrastructure PA market over the next 4-5

years as 5G macro cells with M-MIMO drive a 3-4x content increase relative to a 4G

baseline and the need for smaller mmWave sites in densely, populated areas creates an

entirely new piece of the TAM. Netting it all out, we expect the infrastructure PA TAM to

increase by a factor of 2.1x from the 2018 trough ($1.48B) to the 2024 peak of $3.1B

implying a 13% CAGR. By technology, we expect the LDMOS market to show modest

growth (up >10% from 2020-2024 vs. 2015-2019) while GaN (3-6GHz M-MIMO, some

mmWave) and SiGe (mmWave) drive the majority of the growth in the TAM. Quickly on

the GaN-on-SiC vs. GaN-on-Si, we expect the vast majority (>85%) of GaN deployments

to be GaN-on-SiC-based.

7


First, the argument for GaN-on-Si devices has been that they offer “good-enough”

performance and lower cost relative to GaN-on-SiC devices. However, in our view, it’s

unclear if GaN-on-Si devices actually have a lower total cost relative to GaN-on-SiC

devices. Specifically, the argument for GaN-on-Si is that it leverages the Si substrate

supply chain, which is higher volume and lower cost vs. the SiC substrate supply chain.

The nuance being that SiC substrates have much better power density relative to Si; and

thus, GaN-on-SiC die tend to be much smaller vs. GaN-on-Si.

Second, and more importantly, the GaN-on-Si device supply chain is very immature

relative to the GaN-on-SiC device supply chain. Specifically, Sumitomo (and to a lesser

extent, QRVO) has been shipping qualified products to Huawei, ZTE, and Samsung for

years. Similarly, NXPI and Ampleon are working with their foundry partners (we believe

Win Semi) and are currently sampling GaN-on-SiC parts. Conversely, the legal battle

between MTSI and IFX and ensuing leadership change at MTSI has significantly delayed

the ramp of GaN-on-Si devices. Specifically, MTSI is still working to perfect its process

flow with STM, its high volume manufacturing partner. Also worth noting, Wolfspeed (a

subsidiary of CREE) purchased IFX’s LDMOS PA assets, as well as its ability to sell GaN-

on-Si beyond 2021, but it has instead elected to focus solely on GaN-on-SiC—likely as a

materials supplier to NXPI/Ampleon initially, and an IDM over time as IFX brings back-end

packaging technology and OEM relationships to CREE.

Exhibit 5. RF PA TAM by Application; RF PA TAM by Technology

$M

$500M

$1000M

$1500M

$2000M

$2500M

$3000M

$3500M

2015 2016 2017 2018 2019e 2020e 2021e 2022e 2023e 2024e

MM 4G/5G 5G M-MIMO 5G mmWave

$M

$500M

$1000M

$1500M

$2000M

$2500M

$3000M

$3500M

2015 2016 2017 2018 2019e 2020e 2021e 2022e 2023e 2024e

LDMOS GaAs GAN SiGe

Source: Yole, Mobile Experts, Oppenheimer & Co. Estimates

5G: Massive MIMO & Beam Forming—Key New Technologies

Massive Multiple-Input, Multiple-Output (M-MIMO) is an extension of Multiple-Input,

Multiple-Output (MIMO), a general idea that has been around for decades and has been

used in several wireless communication technologies over the last 5-10 years including

802.11n, 802.11ac, and 4G LTE. To better understand M-MIMO, let’s first understand the

basics of the underlying radio and antenna technology and the progression from Single-In,

Single-Out (SISO) systems to Single-In, Multiple-Out (SIMO) systems, and finally to MIMO

and M-MIMO radio systems.

A single antenna can transmit or receive but not simultaneously; thus historically, wireless

radios were constructed as Single-In, Single-Out (SISO) systems where one antenna sat

at both ends of a link. However, as frequencies became more congested and signal

quality suffered, network architects began to take advantage of the idea of spatial degrees

of freedom. Spatial degrees of freedom can be achieved by adding multiple antennae at

8


the transmitter or the receiver and can be exploited in the form of diversity (i.e.,

redundancy), multiplexing, or both.

Single-In, Multiple-Out (SIMO) transceiver systems were the next step in the evolution.

These systems had one antenna at the transmitter but two at the receiver; thus if the two

receive antennae were trying to receive the same signal, but there was some level of

interference along each path, the two antenna could combine the two signals and

reconstruct one coherent, higher quality signal. The result was much better receive-side

performance vs. SISO systems.

MIMO systems followed SIMO systems and initially also primarily only took advantage of

diversity. In a 2x2 MIMO system, there are two transmit antennae and two receive

antennae. A 2x2 MIMO system would have similar receive performance vs. a SIMO

system but better transmit performance.

However, over time MIMO technology not only scaled from 2x2 to 4x4 or 8x8, but the use

case evolved such that the incremental spatial degrees of freedom were used more for

multiplexing than diversity. Specifically, a 4x4 MIMO system could transmit four unique

streams of information from different antennae, each operating at the same center

frequency, to a corresponding receiver; and thus, transmitting 4x as much data over the

same channel as a SISO system under optimal conditions. Hence, MIMO now most

commonly refers to a practical technique for sending and receiving more than one data

signal out over the same frequency simultaneously by exploiting multipath propagation.

This technique results in a multiplication of the capacity of a radio link and allows it to

serve multiple autonomous terminals (i.e., smartphones) simultaneously over the same

channel.

M-MIMO is simply the continued scaling of MIMO and in technical terms refers to at least

a 16x16 MIMO. That said, thus far in the 5G sub-6GHz launch, we are primarily seeing

32x32/64x64 M-MIMO implementations and significant increases in capacity relative to 4G

LTE. For example, the addition of 3.5GHz 5G RRHs with M-MIMO at an existing 4G

macro site in a dense city resulted in a 6x increase in downlink capacity and a significant

reduction in latency. Further, M-MIMO is an extremely cost efficient method of adding

capacity vs. adding more macro cells as it lowers equipment costs and energy

consumption. Specifically, the array gain permits a reduction in radiated power, and the

combination of low-accuracy signals and linear processing results in excellent

performance and enables a further reduction in power. Secondly, M-MIMO requires L1

processing in the RRH, which in turn drives a 5-10x reduction in FH transmission

demands, all else equal. However, obviously all else is not equal as M-MIMO multiplies

the bandwidth of a particular RRH so FH transmissions still actually increase on an

absolute basis. In turn, most NEMs are using 25G optics for 5G vs. 10G for 4G.

So who wins with M-MIMO? The short answer is everyone, but more specifically, M-MIMO

drives the number of antenna elements and radios per RRH up significantly and, thus, a

proportionate increase in the number of RF and analog signal chain chips required.

Specifically, toward the end of the 4G LTE cycle, most RRHs had either four (4x4 MIMO)

or eight channels (8x8 MIMO), which means four or eight power amplifiers (PAs), low

noise amplifiers (LNAs) and switches, and either one or two transceivers as each

transceiver can handle four channels (4x4). However, as mentioned previously, most FR1

M-MIMO implementations are either 32x32 or 64x64 implying an 8x increase in both RF

and analog signal chain chips. That said, content growth will likely lag unit growth (4-5x vs.

8x) as the cost per channel comes down such that the implementations make economic

sense for the carriers.

Beam Forming (BF) technology is fundamentally different than M-MIMO; however, it is

often used in tandem with M-MIMO implementations. Rather than multiplying the number

9


of data streams that a RU can handle, BF enhances the performance of each single data

stream. Historically, antennas would broadcast a signal in every direction simultaneously

(i.e., 360°). Thus, M-MIMO implementations, which enable the transmission of as many as

256 signals at once, create a significant amount of interference all else equal as the

individual transmissions all interfere with each other. This is where BF technology comes

in—BF essentially transforms a 360° radio transmission into a focused beam, which the

antenna then broadcasts to a specific location or device, thus alleviating the interference

issue. In addition to eliminating interference and, thus, improving the effectiveness of a

transmission, it also enables the delivery of stronger radio signals over greater distances.

This is an ideal solution to increase capacity in congested areas, especially when adding

another frequency band isn’t really an option.

Exhibit 6. FR1 M-MIMO with BF Block Diagram; FR2 M-MIMO with BF Block Diagram

Antenna Antenna

Antenna

Antenna

TransceiverRF

Beamformer

Switch

Switch

RF Beamformer

Transceiver

Source: Qorvo, Analog Devices, Oppenheimer & Co. Estimates

5G: Delivering an Improved Consumer Experience

Looking back, the first generation (1G) network was analog-only and provided support for

voice on mobile phones. The 2G upgrade transitioned the network from analog to digital

and supported text messages in addition to voice calls. The implementation of 3G network

technology provided more capacity and less latency, thus allowing for multi-media support

on mobile devices. Similar to 3G, the 4G upgrade increased capacity, bandwidth. and data

transfer speeds and enabled mobile, wireless internet.

To put it simply, 1G allowed people to use mobile phones, and each transition since then

(from 2G to 3G to 4G) was focused more on data bandwidth and enabled people to do

progressively more with their mobile devices. Not surprisingly, the increase in speed,

bandwidth, and capacity associated with the 5G NR transition will also enable consumers

to do more with their mobile devices. Specifically, the 5G NR standards have been set to

facilitate the adoption of eMBB, which promises to bring features like augmented and

virtual reality (AR & VR) in 4K and 8K resolution to mobile, wireless devices. As such,

Ericsson (ERIC) forecasts that 5G devices will represent 23% of the 8.8B global handset

connections by 2024 but 36% of mobile traffic as 5G subscribers are expected to drive

~2x the amount of traffic vs. 2G/3G/4G subscribers.

10


Exhibit 7. Connected Devices by Technology; Mobile Data Traffic by Device Type

Source: Ericsson, Oppenheimer & Co. Estimates

Further, Cisco’s (CSCO) annual Virtual Networking Index (VNI) report supports the view

that consumers are hungry for eMBB, which promises to bring AR/VR in 4K/8K definition

to smartphones. Specifically, smartphones are expected to represent 25% of global device

connections and 23% of IP traffic in 2019, which suggests IP per smartphone is slightly

lower vs. the average connection. That said, CSCO predicts that 5G (and eMBB) will drive

a +41% CAGR in EB per smartphone connection from 2019-2024, a slight acceleration vs.

the +39% CAGR observed from 2014-2019. Accordingly, EB per smartphone connection

will increase by 5.5x over the next five years, well ahead of the less than 2x growth in EB

per average connection. As such, IP traffic per smartphone connection is expected to be

2.7x higher than the average connection in 2024 vs. slightly below average in 2019

implying smartphones will represent 55% of global IP in 2024 (vs. 23% in 2019) despite

declining to 21% of connections (vs. 25% in 2019).

Exhibit 8. IP Traffic CAGR by Device Type; IP Traffic by Device Type—Indexed to Total

Source: Cisco, Oppenheimer & Co. Estimates

5G: Not Only About the Consumer

While one could argue that the 2G/3G/4G network upgrades largely delivered handset-

oriented improvements for consumers, the 5G build-out also promises to deliver significant

upgrades and/or new use cases to municipalities, governments, and businesses.

Specifically, in addition to eMBB, 5G will support: (1) FWA, which brings high-speed

wireless internet access to homes, enterprises, and/or public gathering points (i.e.,

stadium); (2) URLLC, which will enable industrial application, remote manufacturing, traffic

& safety control, and autonomous driving; and (3) mMTC, which will facilitate connections

between a large number of low-power, low cost IoT type devices across a wide area and

enable features like smart metering, fleet management, and tracking of all sorts.

11


Accordingly, CSCO’s VNI report forecasts a +10% CAGR for global device connections

from 2019-2024, an acceleration vs. the +9% CAGR from 2014-2019. The expected

acceleration is somewhat surprising given that smartphone growth has stalled; however,

global M2M connections (40% of total) already exceed smartphone connections (25% of

total) and are expected to deliver a +18% CAGR from 2019-2024. As such, the M2M

category is expected to drive more than 85% of connection growth over this time frame

and represent ~60% of total connections by 2024—nearly 3x the number of smartphones

connections (~20% of total).

While M2M devices are already the largest single category (40% of total) in terms of

device connections, it’s actually the second smallest category (only larger than feature

phones) in terms of IP traffic. Specifically, the M2M category only represents 4% of global

IP traffic as M2M connections drive 10x less IP traffic vs. an average connection.

However, in addition to driving growth in the number of M2M connections, the advent of

URLLC and mMTC is expected to drive significant growth in the amount of IP traffic per

M2M connection.

Exhibit 9. Connected Devices by Device Type; IP Traffic by Device Type

Source: Cisco, Oppenheimer & Co. Estimates

Specifically, CSCO’s VNI forecasts a +30% CAGR and implies nearly a 4x increase in EB

per M2M connection from 2019-2024. Despite the expectation for the number of EB per

M2M connection to grow at a rate pace over 2x the aggregate, an average M2M

connections will still drive 5x less IP traffic vs. other connections in 2024 (10x less in

2019). As such, the M2M category is still only expected to represent 12% of global IP in

2024 (4% in 2019) despite representing ~60% of connections. That said, it’s worth noting

that we expect the M2M category to continue to increase as a percent of total connections

beyond 2024. Further, we expect M2M as a percent of global IP traffic to increase at an

even faster rate, especially as the adoption of autonomous vehicles drives significant

growth in EB per M2M connection.

Carrier Capex Cycle—5G and Revival of China to

Drive a Reaccelerating Wireless Capex

Global mobile RAN spending ranged from $30B-$35B from 2004-2016 but fell below $30B

beginning in 2017 and has remained there as carriers have slowed spending in an attempt

to increase profits and ahead of a potentially expensive 5G cycle. While we expect the

modest negative long-term trend in the mobile RAN market to persist—at least until

potential new 5G revenue streams for carriers prove out—we do see potential for the

global 5G buildout and reacceleration in wireless spending in China to drive a cyclical

recovery such that the RAN market reverts back above recent trend and toward the $33-

12


$34B range for some time between 2020-2025 implying +15-20% recovery vs. the 2018

trough.

Looking at the global RAN market from a technology perspective, we would note that 2G

marked the peak for the level of spend on a single technology as carriers spent >$25B a

year on 2G from 2004-2007 as there was no 1G network to fall back on (i.e., 3G could

leverage 2G, 4G could leverage 2G/3G) and carriers raced to win subscriber share.

Conversely, dedicated 3G spend never exceeded $21B and dedicated 4G spend never

topped $23B on an annual basis despite them being more expensive technologies vs. 2G

and ramping into much larger subscriber bases.

Peak 3G spend was undoubtedly hindered by: (1) the financial crisis in 2008-2009; and (2)

the fact that 4G initially launched only 7-8 years after the first 3G deployments—

comparatively 4G spend peaked 12 years after initial trials. Aggregate 4G spend has been

much higher vs. 3G ($108B in total over the peak 5-year period vs. $85B for 3G), but

instead of ramping to a peak and falling, 4G spend has essentially plateaued at the $20-

$23B level as the early movers—US, Japan, and South Korea—led the initial ramp and

China lagged behind, thus holding spend at relatively high levels until the slower moving

nations began to shift spend to 4G and offset declines in China, albeit total RAN spend

declined as 2G/3G spend fell in lagging nations.

Exhibit 10. Global Mobile RAN TAM; Global Mobile RAN TAM by Technology

Source: Dell Oro, IHS, Gartner, Oppenheimer & Co. Estimates

Given: (1) the significant growth in the number of frequency bands allocated to carriers

across the globe for 5G NR; (2) the fact that 5G NR will rely on higher frequency/lower

propagation spectrum (and thus shorter reach); (3) the expectation that 5G NR will roll out

in stages with 5G NSA initially and 5G SA to follow (i.e., no 6G to cut cycle short); and (4)

China looks to be one of the early movers in 5G NR, we see the potential for a re-

acceleration in spending as the 5G cycle gains steam and expect annual 5G capex to

likely surpass peak 3G/4G levels and potentially rival the $25B level we saw during 2G

deployments.

While we have discussed many of the potential technical/more sophisticated reasons why

the global RAN market could rebound substantially from current trough levels, let’s look at

something a bit simpler, the power of China—albeit the Huawei ban does create some

question marks with regard to the timing and pace of the China 5G ramp. Specifically,

China Mobile (~60% of mobile subs) received its TD-LTE license and launched

commercial 4G service in February 2014, with China Telecom (~20% of mobile subs) and

China Unicom (~20% of mobile subs) launching commercial service in December 2014

after being granted FD-LTE licenses. As such, total service provider capex in China grew

13


by ~$19.5B (up 41%) from the post 3G-lull in 2012 to the peak of the 4G cycle in 2015

with wireless-only capex up $12.9B (up 69%) over the same time period. During the same

time annual 4G BTS additions increased from basically 0 in 2012 (20k) to >1M in both

2015 and 2016 vs. a normalized global BTS market that ranges from 1.4-1.8M. According

to data from the three Chinese carriers, China represents ~45% of cumulative 4G BTS

deployments and 60-75% of annual deployments during peak years. While the three

carriers spent a lot of money on BTS deployments, they were certainly rewarded in terms

of winning higher value 4G subscribers—12 months after all three carriers had launched,

there were 415M 4G subscribers in China (32% of total subs); and 24 months after

launch, there were 762M 4G subscribers in China (57% of total subs). Today, China has

>1.2B 4G subscribers, which is 30-35% of the total global 4G subscriber base of

approximately 8B.

Exhibit 11. China 4G/5G Base Station Deployments; China Subscribers by Technology

Source: China Mobile, China Telecom, China Unicom

Post the 2015 peak in China service provider spend, total China service provider spend

fell $27.4B (down 41%) to $40.1B and China service provider spend on wireless fell

$14.4B (down 46%) to a trough of $17B. Annual 4G BTS deployments from 2017-2019E

have been about 25% below the peak levels observed in 2015 and 2016. That said, in

total, the three Chinese careers expect to deploy >130k 5G BTS in 2019—China Mobile

>50k, China Telecom >40k, and China Unicom >40k—which is providing green shoots in

terms of service provider spending. Specifically, total service provider spend is expected

to rebound by $2.4B (up 6% Y/Y) to $42.4B and total wireless spend is expected to

recover by $3.2B (up 19% Y/Y) to $20.2B. Note that still implies total service provider

spend of $25.1B or 37% below prior peak and total wireless spend of $11.2B or 36%

below prior peak.

Exhibit 12. China Wireless Capex; China Service Provider Capex

Source: China Mobile, China Telecom, China Unicom

14


Note that up until this point of the section, the expected rebound we have discussed in

carrier capex was purely product cycle and/or cyclically driven; however, if we see new

5G-related applications including Enhanced Mobile Broadband (eMBB), Fixed Wireless

Access (FWA), Ultra Reliable Low Latency Communication (URLLC), or Massive Machine

Type Communication (mMTC) drive significant, sustainable new revenue streams for

carriers, we could see a structural increase in service provider spending and, thus, the

end of the long-term downward carrier capex trend-line.

While it remains to be seen that carriers will be able to drive material business value

above and beyond the temporary premium pricing associated with 5G-enabled eMBB. The

long-term 5G dream is for it to enable service providers to provision specific virtual

networks and/or services and sign service-level-agreements (SLAs) with corporations

across verticals. Further, given the timelines associated with robust, dense, and broad 5G

SA network coverage, we aren’t likely to know if NEMs and carriers will be successful in

realizing this vision until 2025 or later, which is beyond the scope of this report and our

capex forecast.

Basics of a Macro Cell—Base Station, Remote Radio Head, and Antenna in Detail

Wireless infrastructure networks have evolved significantly as technology has both

evolved and improved through the 2G, 3G, and 4G networking upgrade cycles. Further,

5G promises to bring further changes to the network as telecom operators across the

globe work to virtualize their networks in order to make them more agile, boost utilization,

and ultimately lower total cost of ownership (TCO).

As mentioned previously, the shift to SDN and NFV began in the middle of the 4G cycle

and is poised to accelerate with the 5G NSA roll-out and with 5G SA bringing end-to-end

virtualized networks. The transition from solely running specific network tasks on fixed

function appliances to running network functions on industry standard servers in telecom

data centers, as well as corresponding, significant improvements in switching/routing

silicon and optical transport networks, will allow carriers to further evolve and abstract their

wireless RANs; however, for ease of representation, we’ll focus on the more standard

Distributed Radio Access Network (DRAN) architecture vs. any level of Centralized

(CRAN) or Virtualized (VRAN) Radio Access Network in this report.

15


Exhibit 13. Network Block Diagram


Looking more closely at a macro cell, the increased complexity (more spectrum, higher

frequencies, wider bandwidths, and new technologies) associated with 5G is driving

significant increases in content relative to 4G across all three major sections of the cell—

the baseband unit (BBU) or base station (BTS), the radio unit (RU) or remote radio head

(RRH), and the antenna or antenna array.

The BBU sits at the bottom of a macro cell tower and connects up to the RU via a 10/25G

optical front-haul (FH) link. The BBU also connects back to either an aggregation unit or

the network core via a 100/400G optical back-haul (BH) link. The BBU typically has a

modular set-up (similar to a server chassis) and holds at least one main card (MC) and

one line card (LC) —more LCs are typically added to the BBU as more spectrum is

allocated or as next generation technologies roll-out (i.e., LTE vs. LTE-A).

Exhibit 14. Base Station Block Diagram

Source: IDTI, Oppenheimer & Co. Estimates

16


The MC is populated with at least one central processing unit (CPU) or field-

programmable gate-array (FPGA) and handles management, processing, and transport.

The LC is also normally populated with at least one CPU or FPGA (or both), as well as

three to six baseband processors (BB). The LC handles much of the layer 1 (L1), layer 2

(L2), and layer (L3) processing and also handles all of the radio functions (signal

generation, modulation, encoding, and frequency shifting) handled in the BBU. Note, L1 is

the physical layer and refers to the actual hardware components that transmit digital data

across the network, L2 is the data link layer and is used to transmit data between adjacent

nodes in a network, and L3 is the network layer and handles path determination and

logical addressing (IP).

Exhibit 15. 4G to 5G Main and Line Card Diagram

4G 5G Main Card

Line Card

CPU/FPGA

CPU/FPGA

CPU/FPGA

PHY PHY

CPU/FPGA

BB

BB

CPU/FPGA

BB

PHY

BB Switch

BB

CPU/FPGA

BB

PHY

BB Switch

BB

Switch

BB

PHY

Switch

Source: Marvell, Oppenheimer & Co. Estimates

The RU is made up of two primary parts, the digital-front-end (DFE) and the analog-front-

end (AFE) and serves as a go between for the BBU and the antenna or antenna array. In

a transmit situation, the BBU sends a stream of data up to the RU via a FH link. In the RU,

the stream of data first enters the DFE where it is converted into a digital signal. The

digital signal is then passed along to the AFE where it is first transitioned into an analog

signal, and then translated into a RF wave. The RF wave is then blasted up through the

antenna via a fiber to the antenna architecture (FTTA) and out over a frequency (i.e., a

channel) via a power amplifier (PA). In a receive situation, the exact opposite happens.

The antenna receives a generally weak RF signal and passes it to the AFE where it is

initially processed through a low-noise amplifier (LNA). The LNA amplifies and

strengthens the initially weak radio signal while adding relatively little interference (i.e.,

noise) and moves the signal through the rest of the AFE and into the DFE where the

analog to digital conversion occurs. The digital stream is then passed down via the FH link

to the BBU where it is processed.

17


Exhibit 16. Remote Radio Head Block Diagram

Source: IDTI, Oppenheimer & Co. Estimates

The two new technologies associated with the 5G roll-out that are driving significant

content gains in the RU are massive MIMO (M-MIMO) and Beam Forming (BF). In the

following section, we address M-MIMO and BF, as well as how the two technologies

impact semiconductor content in significantly more detail. However, in simple terms, M-

MIMO multiplies the number of antenna, transmit, and receive elements on a given tower

so that the site can transmit and/or receive multiple signals simultaneously via the same

frequency. Given that antennas typically broadcast signals in every direction at once (i.e.,

360°), M-MIMO implementations, which enable the transmission of as many as 256

signals simultaneously, create a significant amount of interference. This is where BF

comes in—BF essentially transforms 360° radio transmissions into focused streams of

data and sends them to a specific location, thus alleviating the interference issue. M-

MIMO and BF are largely analog and RF technologies, but they actually drive significant

content increase across both the AFE and the DFE of the RU.

From an analog and RF perspective, the move to M-MIMO drives the number of radios

per RU up by an average of 8x; thus, the number of transceivers and RF elements

(switch, LNA, PA) increase proportionately. From a digital perspective, the significant

increase in the number of radios, and thus data, increases the amount of processing

power required per RU. As an example, M-MIMO implementations require some level of

L1 processing in the RU as a means of reducing FH transmission demands—note by

putting L1 processing in the RU FH, transmission decreases by 5-10x all else equal;

however, all else is not equal, so the amount of data traveling over a FH link still actually

grows quite significantly, which is why most OEMs are shifting from 10GB with 4G to

25GB with 5G.

18


Exhibit 17. 4G to 5G RRH Diagram

4G 5GRemote Radio Head

DFEASIC/FPGA ASIC/FPGA

AFE AFE

DFEASIC/FPGA

BF ProcessorASIC/FPGA

CPU/FPGA

Source: XLNX, Oppenheimer & Co. Estimates

Macro Cell TAM: BB & RRH Unit and ASP Analysis

Unlike the handset or compute markets, the modular structure of base stations make the

macro cell market a bit more difficult to analyze from a unit and ASP perspective,

especially by technology (i.e., 2G, 3G, 4G, and 5G) as older generation sites can often be

upgraded to newer technologies with software and the addition of line cards. Additionally,

the mid-cycle evolution of radio technologies (i.e., 2G vs. 2.5G, 3G vs. 3.5G and LTE vs.

LTE-Advanced) tends to blur the lines between technologies and further complication unit

analysis by technology. That said, we feel analyzing the macro cell—and thus BTS and

RRH TAM from a unit and ASP perspective—is still quite informative, particularly as it

relates to predicting trends in semiconductor content over time.

First, we’ll look at the landscape from a unit perspective. In aggregate, the macro cell

market has ranged from 1.4-1.9M units annually with individual technologies ramping to

cyclical peaks in the 1.2-1.6M unit range. Looking specifically at the 4G cycle, ~9.2M

macro cells were deployed over the first ten years of the cycle with annual deployments

peaking seven years into the cycle at ~1.4M units. Looking ahead into the 5G cycle, some

industry participants forecast as much as 50% more cumulative 5G deployments relative

to 4G over the first ten years of the cycle and peak annual shipments closer to 2m vs.

~1.5m at the peak of the 4G cycle. The expectation for such strong macro cell growth is

underpinned by the fact that 5G will rely on higher frequency/lower propagation spectrum

(and thus shorter reach), and thus require a higher number of cell sites to cover the same

land area.

19


Exhibit 18. XLNX Analyst Day: Cumulative 4G/5G BTS Deployments; Annual 4G/5G BTS Deployments

Source: Xilinx, Mobile Experts, Oppenheimer & Co. Estimates

While we understand the underlying theory behind these bullish forecasts, we see them as

somewhat optimistic and expect more modest cumulative and peak-to-peak growth with

the two major offsets relative to these more bullish forecasts being: (1) the introduction of

new radio and antenna technologies (i.e., M-MIMO and BF) will improve spectral

efficiency and partially offset the higher frequency/lower propagation spectrum in terms of

Sub-6GHz deployments; and (2) we expect the vast majority of mmWave deployments to

be in the small cell form as FR2 frequencies are so high that macro cells will not be

economical. That said, we forecast 5-15% growth in cumulative and 15-25% growth in

peak-to-peak 5G macro cell deployments relative to 4G. Going a layer deeper, we expect

5G macro BBU growth to approximate our macro cell growth forecast; however, we expect

higher growth in the RRH market as the proliferation of M-MIMO drives the number of

RRH units per macro BTS closer to five on average during the 5G cycle vs. closer to three

on average at the beginning of the 4G cycle.

Exhibit 19. Total Annual Macro Cell Deployments; Macro Cell Deployments by Technology

Source: Mobile Experts, ABI Research, Oppenheimer & Co. Estimates

Next we dig a bit deeper into content trends. In the BBU, we expect 5G adoption trends to

be the primary driver of content gains and expect a 2.5-3.0x increase in 5G BBU content

relative to 4G. More specifically, we expect the adoption of more frequency bands and

wider channel bandwidths to drive both the number of BB chips per line card and the

number of line cards per BBU up by a factor of 2x on average implying a 4x increase in

BB content per BBU. Similarly, we expect the sharp uptick in data plane processing power

and the number of data plane components to drive a roughly 3x increase in control plane

content, and thus we forecast a sharp uptick in CPU content per BBU in 5G relative to 4G.

20


M-MIMO adoption will have some impact on BBU content, but we expect the content

increase associated with the move from 4G to 5G to be much more significant vs. the

move from 5G to 5G with M-MIMO.

Conversely, we view M-MIMO adoption as the biggest driver of content gains in RRHs

given the multiplier effect that an increase in the number of antenna components has on

the RF and Analog signal chain. Note, we expect M-MIMO adoption to be 30-50% through

the 5G cycle but expect higher attach rates, potentially as high as 60-80%, early in the

cycle as initial deployments will be focused on dense, urban areas where M-MIMO is more

of necessity vs. less dense, rural areas, which carriers would target later in the cycle as

they look to expand their coverage footprint. That said, we still expect 5G macro cells

without M-MIMO to have ~1.5x higher RRH content relative to 4G as the number of

radios/channel per RRH still increases, just not to the threshold (16x16) where it’s

considered to be M-MIMO. 5G macro cells with M-MIMO should have ~3x more content

per RRH vs. 4G with 3.5x increase in analog/RF content and a 1.5x increase in DFE

content.

Exhibit 20. BBU Content by Technology; RRH Content by Technology


mmWave: Small Cells Ramp—Again? The concept of small cells was introduced to the carrier networking landscape well over a

decade ago and was initially expected to be somewhat transformational. The idea was

simple, scaled-down macro cells and use these “smaller cells” —there is a range of

different types of small cells (metro, micro, pico, and femto cells), each with different

characteristics in terms of maximum reach and maximum number of devices supported—

to improve density in urban areas where capacity and traffic demands are high but there’s

insufficient space to add a macro cell.

That said, while some small cell deployments have occurred over the last decade—most

popular on campuses, in stadiums, and in office buildings—deployments have been

extremely disappointing in aggregate. In fact, on a unit basis, the macro cell market was

still approximately 3x the size of the small cell market in 2017-2018 vs. initial expectations

for the small cell unit TAM to be measured in the tens of millions vs. the 1.4-1.9M unit

macro cell market.

From our perspective, there have been two primary reasons for the underwhelming

number of small cell deployments to date: (1) most municipalities have regulations

against, and simply don’t want, a large number of smaller cell towers scattered across

roofs or hung from light poles in densely populated areas; and (2) small cells simply

haven’t been economical as the cost to install and maintain small cells doesn’t scale down

proportionately with the lower equipment cost.

21


While issue number one certainly doesn’t go away with the introduction of 5G and remains

a potential gating factor to mmWave adoption, the introduction of mmWave does change

issue number two. In fact, most expect the vast majority of mmWave deployments to be in

the small cell form-factor as macro cells become uneconomical with mmWave frequencies

that only travel a few hundred meters at best vs. Sub-6GHz frequencies that can travel

over thousands of meters.

That said, we do expect small cells to represent an additional tailwind to the mobile RAN

TAM, especially as the adoption of mmWave technology accelerates. Specifically, many

third-party research platforms predict that small cells will increase to 45-55% or total cell

deployments vs. 20-25% at current levels, which suggests an increase in annual small cell

deployments from ~0.5M to ~2.0M (and potentially more) over time. Note, the small cell

revenue TAM will still be much smaller vs. the macro cell TAM as semiconductor content

is typically 5-25x lower in small cells (with analog/RF a higher % of BOM relative to digital

in macro cells) depending on the exact size and type of cell.

The key issue here from our perspective is time—throughout our research and

discussions with industry participants, we found little-to-no enthusiasm for mmWave

technology. Most tend to think it will take 5-7 years before mmWave really hits its stride

as mission-critical use cases, which will require virtualized networks and microwave

technologies, need to be developed. That said, there will be some adoption in 2020,

especially in the US as AT&T and VZ have only been allocated mmWave 5G spectrum.

The reasons for the lack of enthusiasm ranged from: (1) “everything is an issue at

mmWave frequencies” —line-of-sight is required as mmWaves are too wide to travel

through solid objects like trees, walls, and cars, even glass or inclement weather

conditions like snow, rain, or thick clouds will interfere with signals; (2) “it’s a technology in

search of an application” —there’s no real consumer use case and it’s tough to imagine

consumers wanting to pay up for a faster but more spotty/un-even version of Wi-Fi; and

(3) “it’s going to be very hard to put in phone; how do you filter it?” —high-end consumers

could potentially buy dongles, but that’s not a big market, not many people need that

much capacity.

Exhibit 21. 5G Macro Cell vs. Small Cell Deployments

Source: Mobile Experts, ABI Research, Oppenheimer & Co. Estimates

22


Telecom Equipment OEMs

Ericsson (ERIC): Ericsson has been a tier-1 network equipment vendor for decades—

holding over 25% of the Global RAN market each year from 2003-2019E. Given the fact

that almost all initial 5G NR roll-outs are NSA implying the 5G RAN relies on the existing

4G core network, we expect ERIC’s strong position in 4G to serve it well as we progress

further into the 5G NR cycle, as using the same suppliers for 4G and 5G makes it easier

for carriers from an interoperability perspective and creates a level of stickiness for

OEMs. Further, all ERIC radio systems and baseband units manufactured since 2015

support 5G NR and are upgradeable via remote software installation. This strategy

creates another level of stickiness with carriers that used ERIC equipment for 3G/4G

networks and likely gives ERIC a leg-up vs. other OEMs with carriers looking to

modernize their 4G networks and prepare for 5G simultaneously. As such, ERIC has not

reported seeing operators make dramatic shifts from an OEM perspective like it did

during the transition to 4G. ERIC has reportedly signed ~50 commercial agreements for

5G NR. Now looking closer at the first wave of countries moving to 5G NR, ERIC has a

strong position in the US with all major carriers and total RAN share of >40% historically.

ERIC is the close No. 3 player behind Nokia in South Korea, albeit Samsung has a

dominant ~60% market share position 5G cycle to date. In Japan, ERIC only had a

position with SoftBank as NEC and Fujitsu controlled about 30-35% of the market;

however, ERIC expects its position with SoftBank to improve during the 5G cycle and will

have indirect exposure to NTT Docomo, as ERIC is supplying Fujitsu with its BB

processing units. Given China’s preference for domestic suppliers—Huawei, ZTE, and

Datang—ERIC has only had about 10% share in China historically. However, given that

~60% of the world’s 4G BTSs are located in China, ERIC realizes the importance of the

region and has got significantly more aggressive with regard to pricing. China Unicom

recently reported results of a 4G tender, which Huawei won, but ERIC bid 20-25% below

Huawei/ZTE and 40% below NOK. ERIC is hoping aggressive pricing and 5G NR field

trials, which suggest market share gains, will result in an improved overall position in

China. Note the US Huawei ban can only serve to help ERIC in its efforts to build its

footprint in China.

Nokia (NOK): NOK and Alcatel-Lucent (ALU) were both tier-1 network equipment OEMs

with combined global RAN market share peaking at 43% in 2003 and hovering around

40% from 2004-2007. However, the rise of Huawei and the NOK acquisition of ALU in

2016 caused the combined company’s market share to steadily decline and more

recently stabilized in the 22-24% range over the last 3-4 years. Similar to ERIC, NOK

has reported relatively little market share shift amongst carriers moving from 4G to 5G. In

fact, NOK has stated that it has won 5G NR share with every carrier that it worked with

during the 4G cycle (NOK has ~300 4G accounts in total) as 5G NR NSA roll-outs

require the 5G RAN to fall back on the 4G core network, implying that if a carrier were to

introduce a new 5G OEM (prior to the 5G NR SA build-out), the OEM would have to build

a light 4G network in addition to the 5G network. NOK has reportedly signed 45 5G NR

commercial deals and is relatively well represented in the countries that are leading the

transition to 5G NR. Specifically, NOK has a strong position in the US with all major

carriers and total RAN share of >40% historically. In South Korea, NOK has been the No.

1 player historically and held a No. 1 market share position with all three major carriers

(SK Telecom—50% MS is SK; Korea Telecom—30% MS; and LG Uplus—20% MS) for

most of the 4G cycle; that said, NOK appears to be the No. 2 player behind Samsung for

5G NR as Samsung has had a dominant ~60% market share position 5GNR cycle to

date. NOK has been relatively stronger vs. ERIC in Japan historically, albeit only 65-70%

of the market is in play given domestic vendors NEC and Fujitsu control about 30-35% of

the market. Specifically, NOK has always done well with KDDI and NTT Docomo and

recently added SoftBank as a customer for 5G NR. China is by far NOK’s weakest region

with about 10% share and the company fears its position could weaken further through

23


the 5G cycle as China continues to give preference to domestic suppliers—Huawei, ZTE,

and Datang.

Samsung (SCE): Historically, SCE’s RAN business has been subscale and somewhat

overlooked given the conglomerate’s scale in mobile handsets and relative dominance in

both memory and display. That said, Samsung has increased its R&D and focus on the

networks business and appears to be gaining traction. In terms of 4G LTE build-outs,

SCE was a major supplier into South Korea’s large scale deployments throughout the 4G

cycle and began expanding its global footprint later in the 4G cycle with incremental wins

in the US (Sprint, T-Mobile), Japan (KDDI, SoftBank), and India (Reliance Jio). Looking

into the 5G NR cycle, it’s important to note that: (1) SCE was extremely fast to market

with millimeter-wave (mmWave) solutions and the company’s sub-6GHz portfolio is

relatively robust supporting both TDD and FDD technologies, as well as massive MIMO

(M-MIMO); and (2) SCE holds 4G LTE share in three of the four countries that appear to

be driving the 5G capex cycle.

Samsung has 5G tailor-made end-to-end solutions. Late in 4G infrastructure, Samsung

is starting off 5G in South Korea, Japan, and the US where deployments are expected

faster than Europe. South Korea launched the first commercial 5G network, which was

used for FWA at the 2018 Winter Olympics. Nationwide 5G mobile services were

launched in April of 2019, and carrier reports suggests that the country already has >2M

5G subscribers with forecasts suggesting 4-5M by year-end. The network is supported

by 86,000 5G BTS, of which SCE has deployed >60% or 53,000 of those 5G BTS.

Interestingly, the initial SCE 5G BTS roll-out was almost entirely FPGA supported with

essentially no ASIC/ASSPs. However, going forward, Samsung is basically abandoning

FPGAs (more specifically, XLNX) and shifting to MRVL. By our estimates, MRVL content

could top $3300/BTS. Of note, the current 2M South Korean 5G subscribers compare to

a population of 52M and the 86,000 5G BTS compares to 900,000 4G BTS; thus, there’s

a lot of runway for SCE and MRVL in South Korea. SCE has also had commercial 5G

NR success in the US, deploying millimeter wave solutions for AT&T/VZ and is likely also

to participate with S/TMUS given its 4G incumbency. Japan hasn’t transitioned from field

trials to 5G NR scale deployments, but we expect activity to pick up soon ahead of the

Summer Olympics in 2020. Again, we expect Samsung to participate in Japan’s 5G NR

deployments given success with both KDDI and Softbank as the 4G LTE cycle

progressed. Samsung did not participate in China’s LTE deployment in any meaningful

way but has been involved with 5G NR field trials.

Huawei: Huawei has been a dominant force in the global RAN market, steadily

increasing its share from less than 5% in 2006 to become the global market leader for

the first time with 28% share in 2017. Note, Huawei bettered its position and held 31% of

the market in 2018. Perhaps most impressively, Huawei has done this with little to no

exposure to the US due to well-publicized political tension over national security and

privacy between the US and China. Historically, Huawei has been able to overcome its

lack of exposure to the US by leveraging its strong 2G/3G/4G portfolios and aggressive

pricing in emerging geographies such as Eastern Europe, Latin America, Africa, and the

Middle East where Huawei often holds 50% market share or more. Unlike the Western

OEMs, Huawei aligned its 5G NR portfolio with China’s desire to largely forego 5G NSA

in favor of 5G SA and focused on an end-to-end 5G SA solution including core,

transport, RAN, and smartphone. While China has since switched directions and will roll

out 5G NSA initially and the US ban creates further uncertainty, we still expect Huawei to

find a way—bridge inventory, eventually a deal—to participate meaningfully in China’s

5G build-out as China is still Huawei’s most important market (50-55% of revenue in

2018) and Huawei is still China’s preferred OEM (35-45% share with Chinese Telcos).

Looking at the other countries leading the 5G NR charge, both the US and Japan have

elected not to use Huawei equipment. In South Korea, Huawei was only awarded a

minority share position with the country’s smallest operator, LG UPlus—~20% subscriber

24


MS in South Korea. That said, Huawei claims it has 50 commercial contracts and have

shipped 200,000 5G-enabled base stations globally, albeit most likely in China.

ZTE: Telecom equipment is ZTE’s highest priority and most profitable business and

spans customer premise equipment, mobile RAN, transport, and core. In terms of the

global RAN market, ZTE has steadily gained share and has held >10% share from 2014-

2019E. ZTE has diversified its mobile RAN business from a geographic perspective,

such that China is now 45-55% of total depending on the year. That said, ZTE is still

strongest in China where it tends to hold 25-35% of the RAN market vs. 10% globally,

and thus, it created its 5G NR strategy relative to China’s network deployment strategy.

Specifically, China initially planned to largely forego 5G NSA in favor of 5G SA so ZTE

planned to sell carriers on its entire 5G NR solution—RAN, transport, and core. China

has since switched directions and will roll out 5G NSA initially, but we still expect ZTE to

hold a strong position. Both the US and Japan have decided against using Chinese

OEMs for their 5G NR deployments and ZTE was unable to secure any wins with South

Korean operators. That said, ZTE has over 25 5G NR contacts with operators worldwide

and has reportedly shipped over 50,000 BTS across the globe, albeit we suspect the

majority landed in China.

Exhibit 22. Mobile RAN Market by OEM

Source: Dell Oro, IHS, Gartner

Infrastructure

Analog Devices (ADI): Historically, communications infrastructure has represented ~20%

of ADI’s revenue mix with wireless infrastructure representing 65-70% of total segment

revenues implying 13-14% exposure ($800-$850M in CY18—our analysis suggest

HITT/LLTC represent $300-$350M) to the wireless RAN market. The bulk of ADI’s

revenue comes from its analog signal chain power, but contribution from its radio

frequency (RF) via HITT and power management via LLTC portfolios should increase

over time.

ADI is well represented with leading transceiver share across all major global OEMs and

sees potential for its share of analog signal chain content in RRHs to increase from 25-

35% during the 4G cycle to 55-65% during the 5G cycle as OEMs transition to a more

highly integrated transceiver-based architecture with 5G vs. discrete analog ICs

components during the 4G cycle. In terms of content, ADI expects a 4x content increase in

5G Sub-6GHz M-MIMO macro cells and a 5x content increase in 5G mmWave macro

cells relative to a 4G baseline.

25


More specifically, ADI transceivers are 4x4 so during the 4G cycle, a typical base station

would have one (4x4) or two (8x8) transceivers per RRH; however, with 5G M-MIMO most

implementations require either eight (32x32) or 16 (64x64) transceivers implying 8x the

number of chips, albeit only 4x the dollar content as the price of deployment needs to

make sense for OEMs to ramp in volume. Looking more closely at mmWave, ADI expects

roughly a 5x content increase vs. 4G macro cells with the primary content gain coming

from HITT’s beamforming device and up-down converters. Note that while many mmWave

implementations are 256x256, only four transceivers are required (16x16) as the

beamforming happens in front of the transceiver with mmWave implementations vs.

behind it with M-MIMO.

Exhibit 23. ADI Wireless Infrastructure TAM; Global RRH Transceivers Shipments

Source: ADI, Mobile Experts, Oppenheimer & Co. Estimates

In overview, ADI has said that it sees potential for its RAN business to double to $1.6-

$1.7B (>25% growth vs. 2018 baseline) in a flat BTS environment as 5G ramps overs the

next five years. However, in a more bullish scenario, we see potential revenues to

increase by 2.5x to $2.0-$2.1B (32-34% growth), especially as the global RAN market

rebounds cyclically and mmWave ramps into volume.

Marvell Technologies (MRVL): While MRVL has always been strong in networking, it

historically has been relatively weak in the carrier market and was basically absent from

the wireless RAN market. However, the addition of CAVM (and soon Avera) has

strengthened and broadened MRVL’s networking portfolio and provides the company with

significant leverage to the RAN market especially as the 5G NR cycle ramps. Pro-forma

CY18 Networking revenue was $1.5B (~50% of sales), but only $350M (23% of

networking) of it was levered to carrier spend, and the majority was wired vs. wireless.

Specifically, legacy MRVL sold switches, PHYs, and 2-4 core back-haul processors while

legacy CAVM contributed 3G/4G transport CPUs (secured link between BTS and core)

and 4G baseband processors.

CAVM experienced significant content gains moving from 3G to 4G. During the 3G cycle,

CAVM supplied Octeon CPUs for transport security to its lead customer Nokia (per CAVM

10Ks) and had roughly $200 of content per BTS; however, CAVM’s SAM increased 3-4x

to $600-$800 during the 4G cycle as Samsung adopted both CAVM’s Octeon CPUs for

control/data plane function management and Fusion M baseband processors. Looking

ahead into the 5G NR cycle, the increased complexity associated with 5G NR is driving a

near 4x increase to $2,700 in CPU/BB content at Samsung with an incremental $100 from

MRVL switches/PHYs implying $2,800 of total content per BTS. Further, we see potential

content up 5x vs. 4G at Samsung reaching >$3,300 as Samsung is ramping: (1) another

Fusion M processor per RRH (three RRH per macro cell on average during 4G moving to

4-6 with 5G NR) for Massive MIMO implementations—expect 30-40% attach vs. MRVL’s

26


20% estimate—starting this quarter; and (2) a semi-custom chip for front-haul processing

(packet-based link between BTS and RRH) in place of an FPGA in 2HCY21.

Exhibit 24. MRVL: 4G to 5G Transition Drives Higher Content

Source: Marvell, Oppenheimer & Co.

Looking at other OEMs, we expect NOK to ramp its lead 5G platform in the US at

AT&T/VZ using MRVL Fusion M baseband processors ($800-$900 of content) with

potential to replace NOK’s BB ASIC—part of the Reef Shark family of products built by

INTC—for the second- and third-generation platforms given INTC’s manufacturing

struggles with 10nm, especially relatively to an ARM-based process flow. We also note

that it becomes significantly easier for MRVL to win the CPU socket when it is the

baseband incumbent.

Lastly, MRVL’s acquisition of Avera helps get its foot in the door with Ericsson (ERIC) as

Avera has been a digital front-end (DFE) supplier for 15 years. We estimate $165M-

$180M (55-60%) of Avera’s $300M in revenue comes from its digital front-end (DFE) ASIC

business with ERIC and see potential for this business to grow toward $270-$300M during

the course of the 5G cycle given increased complexity and higher number of RRH per cell

site, especially with M-MIMO implementations. We believe Avera has won 7nm 5G DFE at

Ericsson providing a clear sight for revenue stability.

In summary, after taking a deeper dive into the 5G NR market and MRVL’s position within

it, we are raising our base and bull case assumptions relative to potential long-term

incremental revenue/EPS (incl. Avera) from the 5G NR cycle. We now see potential for 5G

to drive incremental ~$1.0B/$0.55E (vs. prior $850M/$0.45E) in our base case scenario

and ~$2.0B/$1.30E (vs. prior ~$1.6B/$1.05) in our bull case scenario. Note that MRVL

recently guided long-term 5G revenue to $600M and said that $1.0B is a somewhat

conservative bull case scenario.

27


Exhibit 25. MRVL: 5G-Driven Revenue & Earnings

Source: Marvell, Oppenheimer & Co.

Qorvo (QRVO): While QRVO is typically thought of as a mobile/smartphone (70-75% of

sales) levered stock or more specifically an AAPL play (35% of sales), we expect the 5G

NR cycle to drive significant top-line infrastructure growth and out-sized bottom-line

growth given the higher-margin nature of the infrastructure & defense (IDP) business. IDP

represents (25-30%) of QRVO sales, and we estimate the wireless infrastructure piece at

approximately $80-120M for CY18 (3-4% of total sales, 10-15% of IDP).

However, during the 3G/4G cycles, QRVO was primarily exposed to mobile RAN via its

small signal (switches/LNAs) business with approximately $100 of content per BTS.

However, during the 5G NR cycle, we expect a 5-6x content gain for small signal devices

in 5G Sub-6GHz M-MIMO BTS plus an incremental 5-6x content gain from its entrance

into the GaN-on-SiC power amplifier (PA) market implying total content of $1,000-$1,200

per 5G BTS M-MIMO.

In small signal, QRVO is relatively well positioned across all OEMs with similar share at all

of the top five players in the market. On the GaN-on-SiC PA side, QRVO does better with

the Asian OEMs—Huawei (albeit Sumitomo is the leader), Samsung, and ZTE as NOK

and ERIC are more comfortable with LDMOS and in combination appear to have been

able to stretch its use case well beyond 2.6-2.8GHz and closer to 3.6GHz. Note that on

QRVO’s C1Q19 earnings call, mgmt. noted >100% Y/Y growth with three of the top five

global wireless OEMs.

Looking a bit more deeply into the content story, we note that as the 4G cycle progressed,

most RRH were either 4-channel (4 LNAs/switches) or 8-channel (8 LNAs/switches);

however, with 5G Sub-6GHz M-MIMO we’re seeing mostly 32x32 or 64x64

implementations implying an 8x multiplier on the number of small signal devices required.

While the chip count is going up by 8x, we expect dollar content for small signal devices to

increase 5-6x to $500-$600 per BTS vs. $100 during the 4G cycle.

Further, the higher frequency bands and wider bandwidths associated with 5G NR

deployments are driving significant adoption of higher-cost GaN-on-SiC PAs—albeit we

expect the LDMOS TAM to be relatively stable—and providing QRVO with a play in the

wireless infrastructure PA TAM, which has been well >$1B for the last 4-5 years and is

likely to double to >$2b over the next 4-5 years. That said, given the lower cost and more

robust LDMOS supply chain, we expect LDMOS to continue to win almost every time at

28


frequencies below 1.8GHz, we expect more of a jump ball with frequencies between 2.1-

3.5GHz and expect frequencies over 3.5GHz to move almost explicitly to GaN. Thus, we

see another $500-$600 of addressable content for QRVO per BTS bringing its total

potential content to $1,000-1,200 per BTS

Exhibit 26. Qorvo 4G/5G Infrastructure TAM

Source: Qorvo, Mobile Experts, Yole, Oppenheimer & Co Estimates

In overview, our analysis suggests QRVO’s mobile RAN TAM will increase by nearly 5x

from ~$400M in 2018 to $1.9B in 2024. While we expect a nearly 5x increase in QRVO’s

revenue TAM, we suspect revenue will increase by something closer to 3-4x given that

~80% of the TAM growth is from GaN-on-SiC PAs where we suspect QRVO will have

lower share vs. its incumbent position in small signal. That said, $80-$120M going to

$240-$480M implies 28-55% growth in IDP and 8-16% total top-line growth.

Texas Instruments (TXN): TXN has made it clear that Industrial (36% of sales) and

Automotive (20% of sales) are its two highest priority end-markets as its internal forecasts

suggest these are the two highest growth, most sticky/stable, and profitable markets. As

such, TXN has increased Industrial and Automotive R&D broadly, and in combination the

two represented 56% of sales in 2018 vs. 42% of sales only five years ago. Relative to

Communications Equipment (11% of sales vs. 15% in 2013), TXN does not expect long-

term growth at a high-level as subscriber saturation and ARPU trending flat-to-down

implies carrier capex is likely to continue on its downward trend. That said, TXN has still

modestly increased its communications equipment R&D budget with initial 5G NR

investments occurring more than ten years ago, but the focus has shifted heavily toward

analog as embedded investments declined.

As noted previously, Communications Equipment represented 11% of TXN sales in 2018

implying the business was $1.7B with 65-70% or $1.1-$1.2B of that designated as

wireless infrastructure revenue. Note that, ~$200M of the $1.1-$1.2B is ASIC business for

ERIC/NOK and is expected to trend to zero over the next 2-4 years implying ~$950M (6%

of sales) of core wireless RAN business. Our analysis suggests that $950M is split roughly

80-85% analog ($760-$810M) vs. 15-20% embedded ($140-$190M). Within analog, TXN

holds a dominant position within power management (PM) and a strong position,

alongside ADI, in the signal chain (SC).

In terms of OEMs, TXN held its strongest market share position with the three largest

global players Huawei, ERIC, and NOK during the 4G cycle but also had a footprint with

Samsung and NEC. Moving forward into the 5G NR cycle, we see potential for some

share loss on the signal chain side as ADI appears to be a bit ahead on transceivers;

however, that should be more than offset by 4x content increases in both the signal chain

29


and power management portion of TXN’s mobile RAN business on 5G NR Sub-6GHz M-

MIMO macro cells. Embedded is likely to trend flat to down slightly, and as previously

noted the $200M of ASICs revenue will decline to zero over the course of the 5G cycle.

Netting it all out, we expect TXN mobile RAN business to stay roughly flat at $1.1-$1.2B

as the 5G cycle progresses.

Intel Corporation (INTC): Intel is best known for its dominance in PCs and servers (and

perhaps its blunders in the smartphone market post its acquisition of IFX’ baseband

business for $1.4B in 2010, since sold to AAPL for $1B in July 2019), but the company

has been quietly building a networking giant through a combination of acquisitions and

internal focus and R&D. Specifically, INTC has seen growth in its networking equipment

business to >$4.0B in 2018 (22% MSS) implying a >40% CAGR vs. $1.2B in 2014 (8%

MSS).

From an M&A standpoint, notable networking-related Intel acquisitions over the last

decade INTC include: (1) Fulcrum Microsystems—Ethernet switch silicon in 2011, (2)

QLGC—InfiniBand assets in 2012, (3) CRAY—HPC Interconnect in 2012, (4) MSPD—

3G/4G BB processors for wireless infrastructure from MTSI in 2013, (5) LSI—Axxia

network processor business from AVGO in 2014, (6) ALTR—FPGA in 2015, (7) eASIC—

structured ASICs in 2018, and (8) Barefoot Networks—Ethernet switch silicon in 2019.

From an internal focus and R&D perspective, INTC entered into long-term

discussions/relationships with carriers nationwide and pitched them on the idea of bringing

the virtualization/flexibility of a hyperscale data center (DC) to carrier networks to lower

TCO vs. the current ridged networks constructed with purpose-built physical appliances

and hindered by low utilization.

Exhibit 27. Core Networking TAM by Technology, Intel Mobile RAN Revenue

Source: Dell’Oro, Intel, Oppenheimer & Co. Estimates

While it’s clear that INTC is committed to networking and has seen phenomenal growth,

we suspect that growth will likely continue for multiple years—the roll-out of 5G NSA and

later 5G SA will significantly accelerate the virtualization of carrier core networks, as well

as drive significant growth in INTC’s currently relatively modest (est. $800-$1B) mobile

RAN business. For this report, we’ll primarily look deeper into INTC’s potential in wireless

RAN as the 5G NR cycle ramps over the next 3-5 years.

Specifically, INTC is targeting 40% of the BTS digital silicon market in 2022, which we

estimate to be ~$4.2B (40% = ~$1.7B), with a range of products including its Snow Ridge

and FPGA combination with ERIC and ZTE and its ASIC-based Reef Shark portfolio of

products—Reef Shark includes BB (in BTS), DFE (in RRH), and an M-MIMO processor (in

RRH)—with NOK. INTC notes that these are very long design cycle products, so it has

high confidence in its 40% MSS target. While INTC hasn’t laid out a specific RRH target

30


MSS, it does have a play in the market with ALTR FPGAs—XLNX expects ~4x content

increase in RRHs for M-MIMO macro cells—and its ASIC-based DFE and M-MIMO

solutions for NOK. In total, we see potential for INTC’s mobile RAN revenue to increase to

$2.45-$2.65B vs. $700M-$800M in 2018 with $1.65-$1.75B from the BTS units and $800-

$900M from RRH units. Note that the $1.8B of incremental growth from 5G NR mobile

RAN implies >45% growth vs. the 2018 networking baseline; however, it only adds 2.5%

of total top-line growth.

NXP Semiconductors (NXPI): Historically, NXPI was primarily exposed to the

communication infrastructure via its LDMOS power amplifier (PA) business. However, it’s

important to note that NXPI sold its business—No. 2 player with 30-35% share—to JAC

Capital (China) for $1.8B in May 2015 to avoid anti-trust issues prior to closing the

Freescale (FSL) deal in December 2015 as FSL, No. 1 player with 55-60% share. We

estimate that RF PA was $700-$800M or 40-45% of NXPI’s $1.8B Communications,

Infrastructure & Other (CI&O) business in 2018 with 80-90% of RF PA exposed to the

mobile RAN market.

In acquiring FSL, NXPI also acquired its Digital & Networking (D&N) business, which was

once a market leader in embedded processing. FSL’s D&N Business peaked in 2010-11

with ~$1.1B in revenue and $300-$400M (30-40%) of exposure to mobile RAN. FSL

primarily provided the embedded CPU for BTS management and transport through the

2G/3G cycle and early in the 4G cycle. FSL was strong with ERIC, ALU, Huawei, and ZTE

but did also have a small position in BB with ZTE. However, given the uncertainty

associated with FSL’s strategic direction (MOT spin-off in 2004, LBO in 2006, IPO in 2011,

and sale to NXPI in 2015) and lack of investment in high-end digital processing, almost all

of its mobile RAN business has transitioned to ASIC (Huawei), INTC (ERIC, NOK, ZTE),

or MRVL (NOK, Samsung). That said NXPI’s current D&N business is roughly $300M,

levered to industrial and enterprise markets, highly profitable, and fairly stable.

Exhibit 28. RF PA TAM by Application; RF PA TAM by Technology

Source: Gartner, Yole, Mobile Experts, Oppenheimer & Co. Estimates

Looking into the 5G NR cycle, we see potential for strong growth in NXPI’s PA business

despite the market transitioning from LDMOS to GaN where NXPI has no exposure (yet).

The four key drivers of growth are: (1) TAM expansion—expect the infrastructure PA TAM

(including multi-market) to more than double to over $3B in 2024 vs. ~$1.5B in 2018. M-

MIMO is a key driver as 5G NR macro cells with M-MIMO implementations tend to have

32x32/64x64 RRHs vs. 4x4/8x8 toward the end of the 4G LTE cycle implying an 8x

increase in PAs. Further, the number of RRHs and antenna sectors per tower looks to be

increasing from ~three on average to ~five during the 5G NR cycle implying up to a 12-

14x PA multiplier. (2) The LDMOS market is likely to be more resilient than most expect; in

fact, we expect a relatively stable LDMOS TAM at ~$1.5B +/- $200M over the next 4-5

31


years as LDMOS continues to dominate below 1.8GHz and holds onto a decent piece of

the market between 2.1-3.5GHz. Further, NXPI’s dominant 55-60% in LDMOS likely

increases as IFX/CREE (10-15% share) focus solely on GaN and Ampleon cedes share at

NOK to NXPI. (3) NXPI ramps production of its GaN-on-SiC PAs and ultimately claims a

strong foothold within the GaN PA market, albeit a weaker position vs. LDMOS. Our work

suggests Win Semi is NXPI’s foundry partner, and we’d highlight that Win Semi has a long

history of supporting successful compound semi businesses—AVGO (cellular PA), Murata

and SWKS (Wi-Fi PA), and ADI/HITT (Infrastructure). (4) NXPI is likely to be very strong

as mmWave solutions ramp given its early mover advantage—already shipping products

into the US (AT&T/VZ) and South Korea—and the combination of performance and cost

that SiGe provides in the 20-40GHz range.

In overview, we see potential for NXPI’s PA business to diversify from 100% LDMOS to

LDMOS, GaN-on-SiC, and SiGe and more than double to $1.6-$1.7B by 2024 vs. a 2018

base-line of $700-$800M implying ~50% CI&O growth and ~10% to total top-line growth.

Maxim Integrated: MXIM has had relatively low exposure to the wireless infrastructure

(~3% of sales) market historically, and we don’t expect a significant change as 5G

technology ramps, as MXIM’s growth strategy is centered around its core positions within

the auto, industrial, and data center markets. MXIM views its general purpose power

management products sold into both BBUs and RRH units as a pure unit play as we move

from 4G to 5G; however, there is potential for share and content gains within MXIM’s

optical product line-up sold into both FH and BH applications. The biggest potentially

driver being the transition from 10G to 25G FH links—MXIM is particularly strong in 25G

as it sells 100G (4x25G) into hyperscale data centers. From an OEM perspective, ERIC

and NOK represent about 80% of MXIM’s wireless infrastructure exposure, with Huawei

and ZTE making up the remaining 20%.

Xilinx: XLNX has historically been over-indexed to the wireless infrastructure market (20-

25% of sales) and expects 5G to be one of its key growth drivers going forward. During

the 4G cycle, 60-65% of XLNX content was in RRHs, 15-20% was in BBUs, and 15-20%

was BH-related. That said, XLNX tends to do better in the BBU early in technology cycles

as smaller OEMs like Samsung and ZTE often initially rely on FPGAs for their complete

BBU platform and shift to an ASIC/ASSP-based approach where they only use an FPGA

as a L1 companion chip as the technology matures. Note larger OEMs tend to take the

ASIC/ASSP plus companion FPGA chip from the start of a cycle.

While the 5G cycle doesn’t seem to be conceptually different, lower BBU content will be a

much more significant headwind for XLNX in 2020 vs. prior years as: (1) dislocations in

ERIC’s (Avera/IBM sold to GF then MRVL) and NOK’s ASIC supply chains (INTC delayed

at 10m); and (2) Samsung’s dominant >60% market share position during South Korea’s

initial 5G build-out represented significant tailwinds for XLNX through 2H18-1H19. XLNX

was the primary BBU silicon supplier for Samsung’s first generation 5G roll-out, but

Samsung is making a wholesale shift to MRVL during its second-generation ramp at the

end of 2019. While lower BBU content will be a headwind in 2020, XLNX expects strong

growth across the rest of its wireless infrastructure portfolio and in total predicts that 5G

will drive a 3-4x increase in revenue relative to 4G. Three key factors underpin XLNX’s

forecast as XLNX expects: (1) cumulative 5G macro cell shipments to exceed 4G by 40-

50% given the lower reach spectrum associated with 5G, (2) FPGA content per 5G macro

cell to increase by 2x relative to 4G as M-MIMO technology proliferates and drives

significant content gains in the RRH, and (3) its share within the FPGA TAM to increase to

~60% during the 5G cycle from ~40% during 4G given its process technology leadership

(shipping 7nm in volume vs. ALTR sampling 10nm) and new product introductions with

RFSoC and Versal. From an OEM perspective, ALTR used to have dominant share within

Huawei and ERIC; however, Huawei’s reliance on FPGAs has declined significantly from

32


peak levels as it shifts to Hisilicon. Further, both Huawei and ERIC are using more XLNX

relative to ALTR vs. through cycle 4G levels.

Infineon Technologies: From a product perspective, IFX is viewed as a best-in-class

supplier of power discrete products with a particularly strong foothold in the automotive

market (40-45% of IFX sales). Historically, IFX has been under-indexed to the

communications infrastructure market (3-5% of sales) and sold its ~$100M RF PA

business (LDMOS/GaN) to CREE in 2018 for $430M. IFX expects the indirect impact of

5G—as an enabler of massive data growth and autonomous driving—to be a bigger

growth driver for the company than the 5G RAN market directly. IFX will also benefit

directly from 5G deployments as 5G: (1) likely reaccelerates macro BBU deployments

where IFX sells AC-DC, DC-DC power; (2) drives the proliferation of M-MIMO, resulting in

a 4x increase in RRH content (from $25 to $100); and (3) results in higher share for IFX

as 5G demands higher-end/quality products where IFX is stronger. From an OEM

perspective, IFX sells to four of the five major global OEMs including ERIC and Huawei.

Semtech: Over the last few years, wireless infrastructure has represented 5-7% of SMTC

revenue depending on timing of deployments and broader end-market trends. SMTC had

leading market share in both the FH and BH market during the 4G cycle and expects 5G

to drive a 3x increase in its revenue TAM relative to 4G, albeit revenue growth is likely to

be somewhat less due to modest share loss from elevated levels. While 5G will drive an

increase in both FH and BH volumes, FH is the primary growth driver for SMTC for 5G as

M-MIMO adoptions drives a significant uptick in FH transmissions. As such, many OEMs

are transition to 25G FH links (from 10G) where SMTC will supply its TriEdge CDR

products, which carry much higher ASPs relative to their FiberEdge PMD products, which

were its high runner during the 4G cycle.

Lattice Semiconductor: LSCC has 40% exposure to Communications Infrastructure and

Computing, and our estimates suggest a rough split of 15-20% Communications and 20-

25% Computing as LSCC improved its content (1.5-2.0x) and attach rate (80% vs. 25%)

on INTC’s Purley sever cycle relative to Grantley. Note, LSCC expects to maintain its 80%

attach rate and benefit from a 2-3x content increase as Whitley roll outs. That said, LSCC

is primarily exposed to the wireless infrastructure market via its position in RRHs. LSCC

sells low-power FPGAs (0.1-1W) into RRHs that are used for control plane functions and

expects a ~30% content gain moving from 4G to 5G. It’s worth noting that LSCC sells its

products into the control plane and doesn’t compete with the much larger, higher-powered

(200W) FPGAs sold by ALTR (INTC)/ XLNX for data plane applications but rather

manages the peripherals around the ALTR/XLNX products in RRHs. From an OEM

perspective, LSCC is fairly well represented across the major OEM with Huawei being the

largest infrastructure customer at 5% of sales.

IPHI: IPHI has significant leverage to the telecom infrastructure market with the segment

representing 30-45% of total sales depending on carrier deployment patterns and relative

strength of cloud data center spend. Specific to telecom, IPHI has historically been

stronger in the transport layer (long-haul/metro) with its Coherent DSPs, TIAs, and drivers

with somewhat lower direct exposure to the RAN part of the carrier cycle; thus, we expect

the transition to 5G SA, which likely ramps in 2022 and beyond, to be a bigger driver for

IPHI relative to the 5G NSA ramp, which is accelerating into 2020. That said, 5G promises

to deliver significantly more capacity/bandwidth relative to 4G and an order of magnitude

increase in FH and BH transmissions. In turn, OEMs are upgrading to higher-performance

optical technologies, which gives IPHI direct exposure to 5G RAN spend via its 50G PAM

DSPs, TIAs, and drivers—historically, more data center-focused products—primarily

through Huawei. In total, IPHI expects PAM4 to go from $120M this year to $320M next

year, albeit the majority of PAM4 revenue (>80% in our estimation) will be driven by cloud

data center vs. service provider spend.

33


Smartphones

Semiconductors 5G will have a profound impact on consumer’s everyday lives. Faster data speeds and

increased connectivity will change the way we interact with wireless devices whether it’s

smartphones, automotive, or IoT devices. In this section of the report, we’ll explore the

smartphone market, the underlying semiconductor enabling technology, and how 5G will:

(1) be a catalyst for content growth, and (2) potentially reaccelerate refresh rates and,

thus, drive incremental unit demand.

LCD: Liquid Crystal Display OLED: Organic Light-Emitting Diode

Handsets represent the largest application in terms of semiconductor demand.

Smartphones are handset devices that run primarily on a high-level operating system that

include but not limited to Android, Apple iOS, Blackberry OS, and Firefox. Smartphones

are generally categorized with features such as a high-quality display (LCD, OLED)

touchscreen, application processor, high-end camera, and the ability to install third-party

applications. According to Gartner, semiconductors for smartphones account for $117B or

24% of the total $475B semiconductor market in 2018.

Exhibit 29. 2018 Semiconductor Market

Source: Gartner

Handsets are a global industry with more than 50 companies competing in the market.

However, the industry is considered somewhat of an oligopoly as the top six OEMs

represent 86% of the $483B smartphone market on a revenue basis and 74% of the1.4B

market on a unit basis. Amongst the top six vendors, Apple, Samsung, and Huawei are

known as tier-1 OEMs and control 51% of the market in unit terms but 71% of the market

in revenue terms given their strength in the premium device market where they leverage

the latest cutting-edge technology. Apple is the single biggest driver behind the

discrepancy between revenue and unit share for tier-1 OEM (37% revenue share, 15%

unit share) as Apple ASPs are nearly 3.5x the global average $255-$260 and nearly 3x

the average of the OEM with the next highest ASP, Samsung at $325. Apple’s market

dominance gives the company influence on the latest handset standards and thus will

prove crucial as the industry transitions to 5G. Conversely, Oppo, Vivo, and Xiaomi are

considered tier-2 OEMs and control 23% of the unit TAM but only 15% of the revenue

TAM.

34


Exhibit 30. 2018 Smartphone OEM Market Share

Source: IDC

The first 5G-enabled smartphones debuted in 1H19 from a few handset OEMs in select

cities and regions around the world. Handset OEMs are making exclusive agreements

with carriers to accelerate 5G roll-outs as each looks to create a first mover advantage. In

turn, this allowed handset OEMs to sell more, higher priced phones and carriers to win

subscriber share by advertising the improved capabilities and services that their 5G

network offers. Handset OEMs are expected to ship 6.5M 5G-enabled devices by the end

of 2019 according to IDC. 5G devices are forecast to grow at a 147% CAGR and have

600M devices shipped in 2024 (37% of total smartphones). We believe Apple, a traditional

laggard in adopting the latest technology, won’t debut a 5G-enabled iPhone until 2H20.

This is strategically justified as 5G infrastructure deployments are still in early stages and

won’t fully ramp in earnest until 2020. With an integrated ecosystem and loyal customer

fanfare, Apple can tap into its 700M iPhone installed base. In comparison, Android

operating phones have a 3.3B installed base. Windows OS account for 5M and other

operating systems (Blackberry, Firefox, Sailfish, HTML-based) for 6.5M of the handset

installed base.

Exhibit 31. Smartphone Shipments and Forecast

Source: IDC, Oppenheimer & Co. Estimates

35


5G Smartphone BOM/Teardown

Samsung sold 292M smartphone devices in 2018, 21% of all smartphone devices sold

worldwide making Samsung the largest handset OEM by unit volume. Samsung is also

one of the first to launch a 5G phone. The Galaxy S10 5G is Samsung’s first phone to

feature 5G capability. A closer look into the bill of materials comparison (Exhibit 32)

between the S10 5G and its close relative the S10+ offers an early view into the content of

first-generation 5G phones and an idea for content potential in future devices.

Exhibit 32. Samsung Galaxy S10+/S10 5G Bill of Materials

Source: Tech Insights

There is a $70 increase in content in the 5G phone. Upon closer look at the component

breakout, two items stand out with the largest content increase: 1) RF Front End and 2)

Baseband Processor. RF content increases by $15 (47% growth) in the 5G phone driven

by the need to address higher complexity from the additional 5G frequency bands. In the

S10+ model, we note the lack of a discrete baseband processor. Since 4G/LTE

technology is mature, now in its 10th year, a combination of technology improvements and

experience, along with intellectual property rights, allow chipmakers to integrate the

baseband into the processor. In the S10 5G model, we see the use of a discrete

baseband. Similar to first-generation 2G/3G/4G devices, we see early 5G handsets

utilizing a discrete baseband as an add-on into existing 4G/LTE-designed devices to

facilitate time-to-market smartphones. The S10 5G uses Qualcomm’s X50 5G modem. On

the new Samsung Galaxy Note 10+ 5G teardown from iFixit (Exhibit 33), note the use of

Qualcomm’s modem and RF modules, highlighting coupling importance between the

baseband and RFFE.

36


Exhibit 33. Samsung Galaxy Note10+ 5G Teardown

Source: ifixit.com

Application Processor

The application processor referred to as the central processing unit (CPU) is the most

critical chip in a smartphone. It is the central hub on a smartphone that performs all

functions, receiving and executing every command. Using an analogy, when compared to

the human body, the processor is considered the “brain.” Every action you perform,

whether it’s opening an image, sending an email, or playing music, gets processed from

binary ones and zeros in the CPU to execute the desired result.

Exhibit 34. Application Processor

Source: IDC, Oppenheimer & Co.

In smartphones, the CPU is integrated as a system on a chip (SoC), which combines

multiple functions including the GPU, baseband processor (modem), security, AI

accelerator, and Bluetooth/WiFi. Apple is the only tier-1 OEM that still uses a discrete

application processor, designing its own A-series chips without a baseband modem. The

industry has moved toward more integrated chips as it reduces space and eliminates the

37


need for two separate chips. Gartner highlights integrated solutions can shave $5 to $7 off

a device’s bill of materials. We believe there will be an increase in demand for discrete

processors in early 5G smartphones to allow handset OEMs entry into the market. As

technology improves, we see the industry moving towards more integrated solutions.

Exhibit 35. Premium Application Processor Chips

Source: Apple, Qualcomm, Samsung, Huawei

Baseband Processor (Modem)

The cellular baseband processor is considered the fundamental technology enabling the

wireless connected world we know today. The baseband referred to as the modem

(abbreviation for modulator-demodulator) is an integrated circuit allowing a smartphone to

connect to the internet and transmit/receive data. The modem converts—“modulates” —a

digital signal into an analog signal to allow transmission over the air through radio

frequencies. Concurrently, it converts analog signals it receives from the antenna into a

digital signal, commonly called data. The baseband works in conjunction with the radio

frequency front-end to manage the cellular network connection. Mobile phones are the

largest market for modems, but adoption is increasing in automotive vehicles and IoT

devices.

Integrated baseband, where the baseband processor is combined with an application

processor, account for 83% of the total baseband market in 2018 based on IDC.

Qualcomm is the largest manufacturer with 57% share in the integrated market, and 53%

of the total baseband market. We believe early 5G phones will use discrete basebands for

initial entry into the 5G market, similar to early 4G/LTE mobile phones in the past. Apple is

currently the only handset OEM that uses discrete basebands, sourcing exclusively from

Intel in 2018 and most recently signing a supply agreement from Qualcomm. We note a

trend in tier-1 OEMs increasing their capabilities to design basebands in-house. Apple

announced plans to acquire Intel’s modem business, joining Samsung and Huawei as

smartphone vendors with capabilities to design basebands and applications processors

internally.

38


Exhibit 36. Discrete Baseband Modems

Source: Qualcomm, MediaTek, Samsung, Huawei, Intel

Application/Baseband Processor Vendors

The market for application and baseband processors share many of the same companies.

Tier-1 handset vendors have capability to design their own applications processors and

baseband, with Apple recently joining the fold. Qualcomm and MediaTek are the two

established incumbent pure play semiconductor companies developing mobile processors

and 5G modems without a handset business. The two account for 60% of the applications

processor market and 66% of the baseband market in 2018, according to IDC. Trends in

the mobile handset market suggest a shift in tier-1 handset OEMs from outsourcing supply

to developing modems in-house for better control chip supply, and less reliance on

Qualcomm’s leading technology, and to mitigate royalty expenses.

Apple is the only major handset OEM that exclusively uses its own internally

developed application processor. The very first iPhone was based on an Apple

SoC manufactured by Samsung. Its first A-series processor was the A4

processor on the iPhone 4. Apple has used Samsung’s foundry for

manufacturing its SoC until the A8, where it transitioned to TSMC and has used

the company ever since. The A13 Bionic is the latest CPU in the A-series lineup

featured discretely on the iPhone 11. Apple signed a 6-year licensing

agreement with Qualcomm (backdated to April 1, 2019), that includes supply of

chips, most likely for early 5G modem. We expect early 5G iPhones will use

discrete Qualcomm basebands. Given Apple’s $1B acquisition of Intel’s modem

business, we believe Apple plans to develop a 5G modem solution and

potentially integrate it within its own process over time.

Qualcomm designs application processors (AP), baseband modems (BB), and

system-on-chip (SoC) solutions, which integrate the AP and BB on a single chip

(Snapdragon SoCs). The BB market has undergone a substantial round of

consolidation over the last 10-15 years as most suppliers exited or sold—ADI,

BRCM, FSL, MRVL, NXPI, STE, and TXN have all shut-down their BB

divisions, Icera was sold to NVDA and later shut-down, and IFX’s BB assets

were sold to INTC and just recently re-sold to AAPL—after realizing they

couldn’t accrue enough scale to support the R&D burden. With most Western

competitors influx and Asian competitors (MTK/SPRD) lagging on technology,

QCOM was left as the only viable 4G BB supplier over the first 2-3 years of the

4G cycle. Further, the OEMs (Huawei, Samsung, ZTE) who had embarked on

internal BB silicon strategies were not ready with 4G product leaving QCOM

39


with substantially all of the 4G market at the time of introduction and >80% for

the next two years. That said, QCOM has lost ~30pts of share over the last five

years (vs. ~50% of the market in 2018) causing QCT revenues to decline 25%

from the peak in 2014 despite smartphone units growing 10-15%. The loss of

AAPL was certainly a significant headwind, but we estimate that QCT is still

down 10-15% ex-AAPL despite benefitting from growth in peripheral markets

(auto, compute, networking, and server) and consolidating the RF360 JV as

both Huawei and Samsung are using significantly more in-house silicon and

MTK has become more competitive in 4G. While we suspect that 5G will be a

tailwind for QCOM, we do not expect the company to regain the dominance it

had at the beginning of the 4G cycle as Huawei and Samsung already have 5G

product ready, MTK is 2-3 quarters behind instead of 2-3 years behind, and

APPL should have internal product ready in the next 2-3 years post acquiring

INTC’s baseband assets this year.

MediaTek is a Taiwan-based semiconductor company that develops CPUs and

baseband modems for mobile devices. Generally considered second tier

compared to Qualcomm in 4G/LTE, the company looks to narrow the

technology gap in 5G modems. MediaTek’s first venture into 5G is with its Helio

M70 5G modem, integrated into a 7nm FinFET SoC. The SoC supports both

NSA and SA 5G architectures but is only capable of supporting sub-6GHz

bands, which keeps complexity and cost of the chips down. MediaTek had 13%

share of the baseband market mostly to Chinese handset OEMs.

Samsung designs and manufactures integrated application/baseband

processors. The company’s Exynos chips are used mostly within its own mid-

and low-tier 4G/LTE smartphones. As 5G proliferates and its technology

advances, we believe Samsung will leverage its manufacturing capabilities to

design chips in-house and reduce reliance on suppliers. Its Exynos 9825 is the

industry’s first processor manufactured on its own leading edge 7nm EUV

technology. Samsung is the only company on this list with fabrication facilities

for high volume chip production. Samsung also manufactures NAND flash

storage and DRAM memory for mobile devices, and is market leader in both

categories.

HiSilicon, a subsidiary owned by Huawei, is a fabless semiconductor company

based in China. The company designs application and cellular baseband

processors for mobile devices. The company claims its Kiren 990 5G processor

is the world’s first SoC integrated with a baseband, its own Balong 5G modem.

HiSilicon’s chips are designed mostly for Huawei smartphone devices though

not exclusively.

Intel (Apple) became an exclusive supplier of 4G/LTE modems to the iPhone

for two years amid Apple’s royalty dispute with Qualcomm. Apple eventually

settled with Qualcomm and reached a 6-year chip supply agreement with a 2-

year extension. Shortly thereafter, Intel announced its decision to exit the 5G

modem business. On July 2019, Apple announced acquisition of Intel’s modem

business for $1B bringing with it 2,200 employees, over 17,000 wireless

patents, and designs for the current XMM 8160 5G modem.

40


Exhibit 37. Baseband Processor Market

Source: IDC

Radio Frequency Front End

The radio frequency (RF) front end module contains all the circuitry between the antenna

and baseband. It processes analog signals received from antennas into a suitable input

for the baseband, while concurrently preparing a digital signal for transmission over radio

waves through its antenna. Due to its proximity with the baseband, the partnership

between RF manufacturers and baseband vendors is critical for designing a power-saving

and performance-enhanced 5G solution. Some of the RF content includes antennas,

power amplifiers, switches, tuners, and filters. 4G RF content trends grows on average

10% annually through the 4G cycle. We expect this trend continues in 5G as complexity in

RF modules for mobile devices increases the need for more content. Since 5G is inherent

in higher frequencies vs. 4G, we expect filters will lead RF content, specifically BAW

filters. By our estimates, we see average RF content growing ~40% from $18 of content

on a 4G handset to $25 in premium 5G smartphones. We estimate a total RF front end

including mmWave market at $26B total by 2025, growth of 8% CAGR. Filters are the

largest sub-component opportunity within RF, a $9.5B market in 2018, growing at an 8.8%

CAGR to $17.1B by 2025. Due to high band counts and rising complexity, BAW filters will

play a bigger role in 5G. A $3.7B market in 2018 (25% of RF content), we estimate it

grows at a 13.4% CAGR to reach a $8.9B TAM (34% of RF content) by 2025.

Exhibit 38. Smartphones RF Market Growth Breakout

Source: Yole, Mobile Experts, Oppenheimer & Co. Estimates

41


Filters

Filters are the largest sub-component and fastest growing within the RF. Radio spectrum

spans from 3Hz to 300GHz. Crowding across 4G/LTE, WiFi, and emerging 5G

frequencies leads to interference and disruptions, leading for the need for more filters. A

filter allows the intended frequency to pass through the front end while rejecting wanted

ones. Filters need to be designed to accommodate various categories of spectrum in 5G,

sub-6GHz, and mmWave, and based on its respective geographic requirements. Band

counts are rising from 13 in 3G, to 45 in 4G, to over 100 in 5G. Filters are the only

component that scales 1-to-1 with the number of frequency bands on a device so we

believe a premium 5G phone could contain up to 100 filters.

Exhibit 39. Expanding RF Content in Mobile; BAW and SAW Market

Source: Skyworks, Mobile Experts, Oppenheimer & Co. Estimates

5G Frequencies

3GPP (3rd Generation Partnership Project) is the governing body that determines protocol

for telecommunication networks. The organization has been involved with the

development and maintenance of 2G, 2.5G, 3G, 4G/LTE, and 5G NR. With Release 15,

frequency bands for 5G NR have been designated. The specification defines two main

bands: 1) FR1—450MHz to 6000MHz (Sub-6GHz), 2) FR2—24GHz to 52.6GHz

(mmWave).

There are 42 frequency bands available in 5G. From a carrier and handset perspective,

they may be interested in providing services at certain bands and channel bandwidths due

to rights (licenses) allocated by the FCC. RF filters would be needed to accommodate

these specific bandwidths. Additionally, there are certain bands that are more attractive

than others. In FR1 (sub-6GHz), interest lies on n77 (3.7GHz), n78 (3.5GHz), and n79

(4.7GHz) as these bands offer wider channel bandwidth up to 100MHz, essential for data

throughput. As for FR2, there are four bands designated for 5G within mmWave spectrum,

n257 (28GHz), n258 (26GHz), n260 (39GHz), and n261 (28GHz) with bandwidth up to

3.2GHz and channel bandwidth up to 400MHz. Hence the need for telecom, handset, and

semiconductor companies to work closely given intermingled complexity of each

participant’s requirements in the wireless ecosystem.

42


Exhibit 40. Select 5G Frequency Bands

Band Frequency Type

FR1 450 to 6000 MHz Sub-6 GHzFR2 24 to 52.6 GHz mmWave

FR1 (Sub-6 GHz)

5G NR Band Band Name Frequency Bandwidth Channel Bandwidth (MHz)

n77 3.7 GHz 3300 - 4200 MHz 900 MHz 10, 20, 40, 50, 60, 80, 90, 100n78 3.5 GHz 3300 - 3800 MHz 500 MHz 10, 20, 40, 50, 60, 80, 90, 100n79 4.7 GHz 4400 - 5000 MHz 600 MHz 40, 50, 60, 80, 100

FR2 (mmWave)

5G NR Band Band Name Frequency Bandwidth Channel Bandwidth (MHz)

n257 28 GHz 26.5 - 29.5 GHz 3 GHz 50, 100, 200, 400n258 26 GHz 24.25 - 27.5 GHz 3.25 GHz 50, 100, 200, 400n260 39 GHz 37 - 40 GHz 3 GHz 50, 100, 200, 400n261 28 GHz 27.5 - 28.35 GHz 850 MHz 50, 100, 200, 400

Source: 3GPP, EverythingRF

Filter Types

There are a variety of RF filters for different devices and markets. The most common

configuration for mobile devices is the acoustic filter. RF filters currently used in the

market for 4G include surface acoustic wave (SAW), temperature compensated SAW (TC-

SAW), incredible high performance SAW (IHP-SAW), bulk acoustic wave (BAW), and thin

film bulk acoustic resonator (FBAR). Given their higher frequency support, BAW filters will

dominate in 5G and will lead RF content growth. We estimate BAW (BAW and FBAR)

market of $3.7B in 2018, grows to $8.9B in seven years, a 13.4% CAGR.

Exhibit 41. RF Front-End Market by Product Type; RF Front-End Market by Air Interface

Source: Yole Developpment, Mobile Experts, Oppenheimer & Co.

Surface Acoustic Wave (SAW) is a compact, low-cost RF filter most suitable for

applications at the low end of the spectrum, for frequencies up to 1.5GHz. It’s

capable up to 3GHz, though performance deteriorates as you approach higher

frequencies, compelling handset OEMs to ideally design them for lower

frequencies. SAW filters are also sensitive at higher temperatures where acoustic

properties are affected, diminishing performance. SAW filters operate by

converting electrical signal energy into an acoustic wave using interdigital

transducers where it travels over a piezoelectric material and then back to an

electrical signal. SAW filters are fabricated on wafers using one or two layers of

43


thin-film metal deposition and one or two photomask (vs. BAW ten mask layers

and many depositions). Thus, SAW filters carry the lowest ASP on this list. A

drawback to SAW filters is they are temperature-sensitive. The stiffness of the

substrate material affects the center frequency of the filter. At lower

temperatures, the frequency shifts upwards. At higher temperatures, the

frequency shifts downwards. To overcome limitations, compensation methods

were developed to reduce sensitivity to temperature changes.

Temperature Compensated SAW (TC-SAW) filters are designed to withstand

sensitivity that SAW filters have at high and low temperatures. TC-SAW filters

are the same as SAW but have an extra layer of coating on IDT structures to

strengthen its stiffness and withstand temperature variation. This reduces the

changes in frequencies. The process requires higher mask layers and raises

manufacturing costs, thereby making its ASP higher than SAW though still less

expensive than BAW filters.

Incredible High Performance SAW (IHP SAW) filters are manufactured by

Murata and look to address some of the technological challenges in traditional

SAW filters and improve their performance. The filters adopt a structure that

makes the energy of surface acoustic wave focus on the surface of the substrate.

This results in filters that exhibit higher Q factor, lower temperature coefficient of

frequency, and improved heat dissipation relative to SAW.

Bulk Acoustic Wave (BAW) filters deliver superior performance vs. SAW and

support higher frequency levels. BAW can address frequencies above 1.5GHz

and up to 6GHz, making them complementary to SAW filters. We believe BAW

will play a bigger role in 5G where early devices will utilize spectrum in the sub-

6GHz frequency range. BAW filter size decreases with higher frequencies

making them ideal in mobile devices where space comes at a premium. Unlike

SAW, the acoustic waves in BAW filters are propagated vertically. BAW filters

have a crystal quartz as the substrate with metal patches on the top and bottom.

The metal patches excite the acoustic waves, which bounce from the top to

bottom. To prevent waves from escaping the substrate, a Bragg reflector is

created by stacking multiple thin layers of alternating materials with varying

stiffness and density. The thickness of the material and mass of the electrodes

determine the frequency where resonance occurs. This method is referred as

solidly mounted resonator BAW (BAW-SMR). BAW filters are less sensitive to

temperature and are more expensive than SAW.

Film Bulk Acoustic Resonator (FBAR) is an alternative form of BAW structured

by having an air cavity beneath the active area. The result is a higher

performance filter with frequency range up to 10GHz. The main difference

between BAW-SMR and FBAR is how acoustic waves are trapped. In FBAR, the

air crystal on both resonators is the primary function that ensures acoustic

energy is trapped. In BAW-SMR, this function is performed by the Bragg

reflector. FBAR filters have higher performance and steeper rejection curves

compared to BAW-SMR and SAW filters. They support frequency ranges from

100MHz to 10GHz making them ideal for 4G/LTE and 5G sub-GHz. FBAR filters

have complex manufacturing challenges, though they provide superior

performance than other filter types, thus carrying the highest ASP on this list. We

estimate ASP of about ~$0.45 per filter. Due to its superior performance, handset

OEMs have chosen to use FBAR on its premium flagship models. Broadcom is

the only supplier shipping FBAR filters in high volume.

44


Exhibit 42. Filter Type

Source: EverythingRF, Murata, EDN

Power Amplifier, Low Noise Amplifier, Switches, Tuners, Diplexer, Transceiver

The RF front-end module contains a sophisticated level of circuity that incorporates

multiple functions between the receiver’s antenna and the baseband, ultimately allowing a

mobile device to transmit and receive data over the air. In addition to filters, there are

supporting components from power amplifiers, switches, tuners, and diplexers that don’t

get as much attention but are just as critical in the RF. In this section, we discuss some of

the other components in the RF module.

Exhibit 43. RF Front End

Source: Qualcomm

45


Power Amplifiers (PA) convert (amplify) low power radio RF signals with data

encoded and modulated at the desired frequency into a higher powered signal for

antenna transmission. The PA sits on the transmit side of the RF signal chain.

GaAs (gallium arsenide) is used mostly today, but at higher frequencies in 5G,

GaN (gallium nitride) will become more attractive semiconductor compound.

Low Noise Amplifier (LNA) is similar to a PA but is positioned at the receiver

channel of the front end. It takes the weak low-power, low-voltage signal received

from the antenna and amplifies it to the desired level without adding distortion or

noise. The signal then gets sent to the RF transceiver.

Antenna Tuners are used to recover performance loss that comes from reduced

efficiencies in antennas. Antenna performance is reduced from challenges in: 1)

industry trends toward smaller antennas, and 2) need for more antennas to

support high data rates (carrier aggregation, Wi-Fi, GPS, 5G). Aperture tuning

and Impedance tuning are two tuning methods used to recover some of this loss

performance.

RF Switches are components used to route (switch) signal paths of high

frequency circuits through transmission paths in a wireless devices. There are

multiple RF switches in a smartphone for various functions. The most common

application is the primary antenna transmit/receive (Tx/Rx) switch, used to

connect the main antenna to either the transmission or reception function.

Diversity Switch is a type of RF switch. Growing demand for 4G/LTE and 5G

band support as well as non-cellular services (WiFI, GPS, Bluetooth) presents a

challenge for maintaining high data integrity. To alleviate burden and support for

multimode/multiband functions, smartphones have a dual antenna system. A

primary antenna is the main antenna that performs Tx/Rx functions while a

secondary diversity antenna does Rx only. The diversity switch enables support

between the two antennas, resulting in higher data quality and reliability.

Diplexer is a simplest form of a multiplexer. A diplexer combines two different

frequency bands, one from the receive path and the other in the transmit path,

into one path. Conversely, it can act as a splitter to enable the signal from one

path to split into two separate paths. The two frequency bands are usually far

apart in frequencies for a diplexer to perform ideally.

RF Transceiver houses both the transmitter and receiver modules in a single

package. The device is located between the baseband and PA/LNA. The

transmitter (Tx) side takes modulated data from the baseband and up-converts

into a RF signal suitable for transmission through the antenna. The receiver (Tx)

does the inverse operation. It takes RF signals received and down-converts to a

data for demodulation at the baseband.

46


Exhibit 44. RF Transceiver Market; Power Amplifier Market

Source: Akoustis, Oppenheimer & Co.

RF Front-End Vendors The RF front-end market is dominated by three incumbent leaders—Broadcom, Skyworks,

and Qorvo. Technology shift into 5G opens the landscape for new entrants each targeting

various opportunities. Qualcomm is looking to provide a complete modem-antenna

solution, whereas Akoustis and Resonant are innovating new techniques to take share in

the $40B RF market. Knowles is one of the few companies that are developing a

mmWave filters, albeit starting at the infrastructure side.

Broadcom is a formidable company in 4G/LTE with a broad portfolio of front-

end modules and looks to extend its lead in 5G with its leading FBAR filters. Its

modules include filters, power amplifier, and multiplexer components, but its

competitive advantage come from its leading and industry first commercial

FBAR filter. Due to its superior ability to function more effectively at congested

spectrum, Broadcom’s FBAR gain significant market share in the handset

market in 4G/LTE. With high barrier to entry from innovative design and high

manufacturing costs, Broadcom sees little competition to no competition at high

frequencies. Its FBAR filters are designed into Apple and Samsung flagship

premium smartphones. Broadcom signed a two-year supply agreement to

provide RF components to Apple in June ’19.

Skyworks has worked with 3GPP since 2015 to develop 5G standards. It has a

portfolio of RF components including amplifiers, switches, tuners, and filters.

The company is best known for its strong position in SAW and TC-SAW filters

after fully purchasing its JV from Panasonic in August 2016. Limited in BAW

and in preparation for 5G, Skyworks is investing to develop BAW in-house.

Expect 35-45% handset content gain, going from $18 content in 4G to $25 per

device in 5G. Core 2G/3G/4G RF systems remain mostly the same, going from

$18 to $20, with sub-6GHz 5G adding another $5.

Qorvo is a broad supplier of RF components to handset device. Formed from

the merger of TriQuint and RFMD, Qorvo is a broad supplier of RF front-end

components. Higher complexity in 5G drives content at high-end handset from

$18 to $25. Its RF Fusion integrates PA, transmit/receive, and switch solutions

into a single package to enhance performance and reduce size. Its portfolio of

modules serves the high-band, mid-band, and low-band at cellular connection.

47


Qualcomm is a leader in the application and baseband processor markets

where industry conditions (OEM in-sourcing, competitive pricing) make it

difficult to achieve sustained growth. In an attempt to bolster growth, QCOM is

hoping to use the introduction of 5G technology to expand its presence in the

RFFE market—primarily limited to envelope tracking solutions during the 4G

cycle—by bundling its 5G baseband with its portfolio of RF products and

offering complete modem-to-antenna solutions. We expect this strategy to be

somewhat successful early in the 5G cycle, especially at tier-2 and tier-3 OEMs

without internal silicon ventures, but our checks suggest most OEMs prefer to

work with their existing RFFE supplier base and plan to revert back to traditional

RF players as alternate baseband solutions come to market. It’s worth noting

that QCOM has been unsuccessfully attempting to make a splash in the RFFE

for roughly a decade. QCOM initially hoped to transform the market with its low-

cost but “good-enough” CMOS-based PAs but never gained significant traction

as performance was never quite “good enough” relative to incumbent GaAs

solutions. QCOM transitioned away from CMOS in the middle of the 4G cycle

and now works with its foundry partner, Win Semiconductor, on GaAs-based

PAs. QCOM doubled down on its RF efforts and added in-house filter design

and manufacturing capabilities to its portfolio by entering into a joint venture

with TDK Corporation. Historically, 51% of the JV was owned by Qualcomm

and 49% by EPCOS (a subsidiary of TDK); however, QCOM recently exercised

its right of first refusal and acquired the remaining TDK shares for $1.15B,

valuing the entire JV as $3.1B. QCOM now has the capability to offer complete

end-to-end solutions with the SnapDragon 5G Modem-RF system, which

includes 5G sub-6 and mmWave modem, PAs, filters, multiplexers, antenna

tuning, LNAs, switching, and envelope tracking.

Murata is a Japan-based electronic components manufacturer that specializes

in the design and manufacture of ceramic filters, high-frequency parts, and

sensors. The company specializes in IHP-SAW filters, which has a limited

market, though it looks to position in 5G with BAW. Murata invested $7M out of

the $10M equity raise from Resonant. As part of the agreement, Muranta gains

access to Resonant’ s filter design technology.

Akoustis plans to make a mark in 5G handsets with its proprietary single

crystal XBAW technology. The company’s XBAW process enables development

of RF filters in the 1 to 7 GHz frequencies that are 23x smaller than legacy

dielectric resonators. Its value proposition includes: 1) wider band support; 2)

improved power handling; and 3) high acoustic performance than incumbent

solutions. Akoustis is positioning for the large handset market but is also

leveraging its tech for development in WiFi and base stations.

Resonant aims to address development of filters through a software approach.

The company’s software platform, Infinite Synthesized Networks (ISN), looks to

design filters through software simulation rather than current processes that are

through iteration. The goal of ISN is to reduce development time and open

access to designing new filter types more cost effectively. ISN is compatible for

SAW, TC-SAW, and recently into BAW, which they call XBAR.

48


Knowles is an early entrant developing a filter solution to address the set of

new challenges in mmWave. Current market filters are applicable for sub-6GHz,

but mmWave requires a new type of filter. Cavity, Planar Thin Film, and

Waveguide are three candidates that support coverage in mmWave frequency

range. Many criteria need to be evaluated, but at this stage, Knowles notes

Planar Thin Film filters is a leading candidate due to its smaller size, cost

advantage, and performance.

Exhibit 44. Filter Spectrum Coverage

Source: Akoustis, Yole, Oppenheimer & Co. Estimate

49


Acronyms 3GPP Third Generation Partnership Project 5G NR 5G New Radio ASIC Application Specific Integrated Circuit ASP Average Selling Price BAW Bulk Acoustic Wave BBU Baseband Unit BF Beam Forming BTS Base Station CAGR Compounded Annual Growth Rate CPU Central Processing Unit DFE Digial Front End EB Exabyte eMBB Enhanced Mobile Broadband FBAR Film Bulk Acoustic Resonator FDD Frequency Division Duplex FWA Fixed Wireless Access GaAs Gallium Arsenide GaN Gallium Nitride GaN-on-SiC Gallium Nitride on Silicon Carbide GHz Gigahertz IHP Incredible High Performance KT Korea Telecom LCD Liquid Crystal Display LDMOS Laterally Diffused Metal Oxide Semiconductor LG Uplus LGU LNA Low Noise Amplifier LTE Long Term Evolution M2M Machine to Machine MHz Megahertz MIMO Multiple-In, Multiple-Out mMTC Massive Machine Type Communication NEM Network Equipment Manufacturers NFV Network Function Virtualization NSA Non-standalone OEM Original Equipment Manufacturer OLED Organic Light Emitting Diode PA Power Amplifier RAN Radio Access Network RF Radio Frequency RRH Remote Radio Head SA Standalone SAW Surface Acoustic Wave SDN Software Defined Networking SiGe Silicon Germanium SIMO Single-In, Multiple-Out SISO Single-In, Single-Out SKT SK Telecom SLA Service Level Agreements SoC Systems on a Chip TCO Total Cost of Ownership TDD Time Division Duplex Tx/Rx Transmit/ Receive URLLC Ultra Reliable Low Latency Communication

50


Stock prices of other companies mentioned in this report (as of 10/1/19): Ericsson Class B (ERIC.B-SE, €78.44, Not Covered) Nokia (NOKIA-FI, €4.49, Not Covered) Samsung Electronics Co., Ltd. (005930-KR, KRW48850, Not Covered) ZTE Corporation Class H (763-HK, HK$20.8, Not Covered) Xilinx, Inc. (XLNX, $92.04, Not Covered) Infineon Technologies AG (IFX-DE, €16.36, Not Covered) Lattice Incorporated (LTTC, $0.01, Not Covered) Inphi Corporation (IPHI, $60.24, Not Covered) MediaTek Inc (2454-TW, NT$376.5, Not Covered) Murata Manufacturing Co., Ltd. (6981-JP, ¥5328, Not Covered) Resonant, Inc. (RESN, $2.73, Not Covered) Knowles Corp. (KN, $19.9, Not Covered) SK Telecom Co., Ltd. (017670-KR, KRW239000, Not Covered) LG Electronics Inc. (066570-KR, KRW66500, Not Covered) MACOM Technology Solutions Holdings, Inc. (MTSI, $20.5, Not Covered) China Telecom Corp. Ltd. Class H (728-HK, HK$3.57, Not Covered) China Mobile Limited (941-HK, HK$64.85, Not Covered) China Unicom (Hong Kong) Limited (762-HK, HK$8.32, Not Covered) Cray Inc. (CRAY, $35.01, Not Covered) International Business Machines Corporation (IBM, $143.66, Not Covered)

51


Disclosure AppendixOppenheimer & Co. Inc. does and seeks to do business with companies covered in its research reports. As a result,investors should be aware that the firm may have a conflict of interest that could affect the objectivity of this report.Investors should consider this report as only a single factor in making their investment decision.

Analyst Certification - The author certifies that this research report accurately states his/her personal views about thesubject securities, which are reflected in the ratings as well as in the substance of this report. The author certifies that no partof his/her compensation was, is, or will be directly or indirectly related to the specific recommendations or views containedin this research report.Potential Conflicts of Interest:Equity research analysts employed by Oppenheimer & Co. Inc. are compensated from revenues generated by the firmincluding the Oppenheimer & Co. Inc. Investment Banking Department. Research analysts do not receive compensationbased upon revenues from specific investment banking transactions. Oppenheimer & Co. Inc. generally prohibits anyresearch analyst and any member of his or her household from executing trades in the securities of a company that suchresearch analyst covers. Additionally, Oppenheimer & Co. Inc. generally prohibits any research analyst from serving as anofficer, director or advisory board member of a company that such analyst covers. In addition to 1% ownership positions incovered companies that are required to be specifically disclosed in this report, Oppenheimer & Co. Inc. may have a longposition of less than 1% or a short position or deal as principal in the securities discussed herein, related securities or inoptions, futures or other derivative instruments based thereon. Recipients of this report are advised that any or all of theforegoing arrangements, as well as more specific disclosures set forth below, may at times give rise to potential conflictsof interest.

Important Disclosure Footnotes for Companies Mentioned in this Report that Are Covered byOppenheimer & Co. Inc:Stock Prices as of October 2, 2019Analog Devices (ADI - NASDAQ, $109.78, OUTPERFORM)Marvell Technology Group (MRVL - NASDAQ, $24.00, OUTPERFORM)Texas Instruments (TXN - NYSE, $128.59, OUTPERFORM)Intel Corp. (INTC - NASDAQ, $50.76, PERFORM)Maxim Integrated Products (MXIM - NASDAQ, $56.83, PERFORM)Semtech Corp. (SMTC - OTC, $47.34, OUTPERFORM)Apple Inc. (AAPL - NASDAQ, $224.59, PERFORM)QUALCOMM Incorporated (QCOM - NASDAQ, $75.47, PERFORM)Broadcom Ltd. (AVGO - NYSE, $274.85, OUTPERFORM)Skyworks Solutions, Inc. (SWKS - NASDAQ, $77.41, PERFORM)Qorvo, Inc. (QRVO - NASDAQ, $73.17, PERFORM)Akoustis Technologies (AKTS - NASDAQ, $7.58, OUTPERFORM)Cree, Inc. (CREE - NASDAQ, $49.22, PERFORM)Cisco Systems (CSCO - NASDAQ, $47.74, OUTPERFORM)NXP Semiconductors NV (NXPI - NASDAQ, $108.97, PERFORM)Sprint (S - NYSE, $6.16, NOT RATED)AT&T, Inc. (T - NYSE, $37.41, OUTPERFORM)T-Mobile (TMUS - NASDAQ, $78.20, PERFORM)Verizon (VZ - NYSE, $59.85, OUTPERFORM)NVIDIA Corp. (NVDA - NASDAQ, $174.00, OUTPERFORM)Universal Display Corp. (OLED - NASDAQ, $165.41, PERFORM)

All price targets displayed in the chart above are for a 12- to- 18-month period. Prior to March 30, 2004, Oppenheimer & Co.Inc. used 6-, 12-, 12- to 18-, and 12- to 24-month price targets and ranges. For more information about target price histories,

52


please write to Oppenheimer & Co. Inc., 85 Broad Street, New York, NY 10004, Attention: Equity Research Department,Business Manager.

Oppenheimer & Co. Inc. Rating System as of January 14th, 2008:

Outperform(O) - Stock expected to outperform the S&P 500 within the next 12-18 months.

Perform (P) - Stock expected to perform in line with the S&P 500 within the next 12-18 months.

Underperform (U) - Stock expected to underperform the S&P 500 within the next 12-18 months.

Not Rated (NR) - Oppenheimer & Co. Inc. does not maintain coverage of the stock or is restricted from doing so due toa potential conflict of interest.

Oppenheimer & Co. Inc. Rating System prior to January 14th, 2008:

Buy - anticipates appreciation of 10% or more within the next 12 months, and/or a total return of 10% including dividendpayments, and/or the ability of the shares to perform better than the leading stock market averages or stocks within itsparticular industry sector.

Neutral - anticipates that the shares will trade at or near their current price and generally in line with the leading marketaverages due to a perceived absence of strong dynamics that would cause volatility either to the upside or downside, and/or will perform less well than higher rated companies within its peer group. Our readers should be aware that when a ratingchange occurs to Neutral from Buy, aggressive trading accounts might decide to liquidate their positions to employ the fundselsewhere.

Sell - anticipates that the shares will depreciate 10% or more in price within the next 12 months, due to fundamental weaknessperceived in the company or for valuation reasons, or are expected to perform significantly worse than equities within thepeer group.

Distribution of Ratings/IB Services Firmwide

IB Serv/Past 12 Mos.

Rating Count Percent Count Percent

OUTPERFORM [O] 397 65.30 191 48.11

PERFORM [P] 210 34.54 68 32.38

UNDERPERFORM [U] 1 0.16 0 0.00

Although the investment recommendations within the three-tiered, relative stock rating system utilized by Oppenheimer & Co.Inc. do not correlate to buy, hold and sell recommendations, for the purposes of complying with FINRA rules, Oppenheimer& Co. Inc. has assigned buy ratings to securities rated Outperform, hold ratings to securities rated Perform, and sell ratingsto securities rated Underperform.Note: Stocks trading under $5 can be considered speculative and appropriate for risk tolerant investors.

Company Specific DisclosuresOppenheimer & Co. Inc. makes a market in the securities of AAPL, ALTR, BRCM, CELL, CRAY, CREE, CSCO, ERIC, HITT,IDTI, INTC, MRVL, MSPD, MXIM, NVDA, NXP, OLED, QCOM, QRVO, SIMO, SMTC, SWKS, T and TMUS.

In the past 12 months Oppenheimer & Co. Inc. has provided investment banking services for AKTS.

In the past 12 months Oppenheimer & Co. Inc. has managed or co-managed a public offering of securities for AKTS.

In the past 12 months Oppenheimer & Co. Inc. has received compensation for investment banking services from AKTS.

53


Oppenheimer & Co. Inc. expects to receive or intends to seek compensation for investment banking services in the next 3months from CREE, HITT, IDTI, MSPD and OLED.

Additional Information Available

Company-Specific Disclosures: Important disclosures, including price charts, are available for compendiumreports and all Oppenheimer & Co. Inc.-covered companies by logging on to https://www.oppenheimer.com/client-login.aspx or writing to Oppenheimer & Co. Inc., 85 Broad Street, New York, NY 10004, Attention: Equity ResearchDepartment, Business Manager.

Other DisclosuresThis report is issued and approved for distribution by Oppenheimer & Co. Inc. Oppenheimer & Co. Inc. transacts business on all principalexchanges and is a member of SIPC. This report is provided, for informational purposes only, to institutional and retail investor clients ofOppenheimer & Co. Inc. and does not constitute an offer or solicitation to buy or sell any securities discussed herein in any jurisdictionwhere such offer or solicitation would be prohibited. The securities mentioned in this report may not be suitable for all types of investors.This report does not take into account the investment objectives, financial situation or specific needs of any particular client of Oppenheimer& Co. Inc. Recipients should consider this report as only a single factor in making an investment decision and should not rely solelyon investment recommendations contained herein, if any, as a substitution for the exercise of independent judgment of the merits andrisks of investments. The analyst writing the report is not a person or company with actual, implied or apparent authority to act on behalfof any issuer mentioned in the report. Before making an investment decision with respect to any security recommended in this report,the recipient should consider whether such recommendation is appropriate given the recipient's particular investment needs, objectivesand financial circumstances. We recommend that investors independently evaluate particular investments and strategies, and encourageinvestors to seek the advice of a financial advisor. Oppenheimer & Co. Inc. will not treat non-client recipients as its clients solely by virtueof their receiving this report. Past performance is not a guarantee of future results, and no representation or warranty, express or implied,is made regarding future performance of any security mentioned in this report. The price of the securities mentioned in this report andthe income they produce may fluctuate and/or be adversely affected by exchange rates, and investors may realize losses on investmentsin such securities, including the loss of investment principal. Oppenheimer & Co. Inc. accepts no liability for any loss arising from theuse of information contained in this report, except to the extent that liability may arise under specific statutes or regulations applicable toOppenheimer & Co. Inc. All information, opinions and statistical data contained in this report were obtained or derived from public sourcesbelieved to be reliable, but Oppenheimer & Co. Inc. does not represent that any such information, opinion or statistical data is accurate orcomplete (with the exception of information contained in the Important Disclosures section of this report provided by Oppenheimer & Co.Inc. or individual research analysts), and they should not be relied upon as such. All estimates, opinions and recommendations expressedherein constitute judgments as of the date of this report and are subject to change without notice. Nothing in this report constitutes legal,accounting or tax advice. Since the levels and bases of taxation can change, any reference in this report to the impact of taxation shouldnot be construed as offering tax advice on the tax consequences of investments. As with any investment having potential tax implications,clients should consult with their own independent tax adviser. This report may provide addresses of, or contain hyperlinks to, Internet websites. Oppenheimer & Co. Inc. has not reviewed the linked Internet web site of any third party and takes no responsibility for the contentsthereof. Each such address or hyperlink is provided solely for the recipient's convenience and information, and the content of linked thirdparty web sites is not in any way incorporated into this document. Recipients who choose to access such third-party web sites or followsuch hyperlinks do so at their own risk.This research is distributed in the UK and elsewhere throughout Europe, as third party research by Oppenheimer Europe Ltd, whichis authorized and regulated by the Financial Conduct Authority (FCA). This research is for information purposes only and is not to beconstrued as a solicitation or an offer to purchase or sell investments or related financial instruments. This research is for distribution onlyto persons who are eligible counterparties or professional clients. It is not intended to be distributed or passed on, directly or indirectly, toany other class of persons. In particular, this material is not for distribution to, and should not be relied upon by, retail clients, as definedunder the rules of the FCA. Neither the FCA's protection rules nor compensation scheme may be applied. https://opco2.bluematrix.com/sellside/MAR.actionDistribution in Hong Kong: This report is prepared for professional investors and is being distributed in Hong Kong by OppenheimerInvestments Asia Limited (OIAL) to persons whose business involves the acquisition, disposal or holding of securities, whether as principalor agent. OIAL, an affiliate of Oppenheimer & Co. Inc., is regulated by the Securities and Futures Commission for the conduct of dealingin securities and advising on securities. For professional investors in Hong Kong, please contact [email protected] for all mattersand queries relating to this report. This report or any portion hereof may not be reprinted, sold, or redistributed without the written consentof Oppenheimer & Co. Inc.

54


https://opco2.bluematrix.com/sellside/MAR.action

https://opco2.bluematrix.com/sellside/MAR.action

mailto:[email protected]

This report or any portion hereof may not be reprinted, sold, or redistributed without the written consent of Oppenheimer & Co. Inc.Copyright © Oppenheimer & Co. Inc. 2019.

55


Documents

5G—Next-Generation Wireless KEY POINTS