Philips tech. Rev. 40, 99-111,1982, NoA 99

The TRAP ATT oscillator

R. Davies, B. H. Newton and J. G. Summers

It is not uncommon in the history of a new semiconductor device for many years to elapsebetween. the first reports of exciting research results and the eventual emergence of a maturecomponent. The TRAPATT diode is no exception, and the promise of ten years ago is onlynow reaching fulfilment. This diode can provide high peak powers at microwave frequenciesof several GHz, but careful circuit design is necessary to capitalize on the full capabilitiesof the device. Research scientists at Philips Research Laboratories, Redhill, and developmentengineers at Mullard Ltd have cooperated in a systematic experimental investigation ofTRAPATT diodes and the oscillator circuits in which they function. As a result, theTRAPATT oscillator can now be considered as a serious contender for pulse-transmitterapplications in the frequency range 2-4 GHz.

Introduetion

Since the early sixties, many solid-state diodes havebeen studied as potentially useful sources of micro-wave energy. Sometimes these devices have been usedas low-power oscillators requiring separate externalamplification to achieve useful power levels. How-ever, many studies have focused on their character-istics as fundamental microwave oscillators producingsufficient power directly. Both continuous and pulsedoperation have been Investigated.The IMPATT (IMPact Avalanche and Transit

Time) diode [1] was the subject of considerableresearch over a period of about ten years, starting in1965. This is a p-n junction diode that has a negativeresistance when reverse biased; its operation dependson the' transit time of carriers generated in an ,ava-lanche region passing through the device. Duringstudies of this diode, a mode of operation wasobserved that differed markedly from the normalIMPATT mode. This new behaviour was character-ized by its very high electrical efficiency, high poweroutput and lower frequency of oscillation, and wasreferred to as the 'anomalous mode' [2].

Out of this work was developed the TRAPATT-diode oscillator, where the acronym 'TRAPATT'stands for TRApped Plasma Avalanche TriggeredTransit. This is the name by which the anomalous

R. Davies, Ph.D., B. H. Newton, Ph.D., and J. G. Summers,B.Sc., are with Philips Research Laboratories (PRL), RedhilI,Surrey, England.

mode is now known, and describes the operation ofthe diode when incorporated in a suitable circuit.Several computer simulations have successfully ex-plained the basic TRAPATT mechanism [3], but nodetailed consideration has been given to the inter-action between the diode and the circuit. Consequent-ly, an empirical approach to device and oscillator de-sign has been followed that, until recently, has led tosub-optimal designs and over-complex circuits, givingrise to unreliable operation.

Initial experimental results using this new type ofdiode were very encouraging, and high peak powerswere obtained at frequencies up to about 10 GHz,with d.c. to r.f. conversion efficiencies greater than300/0. The present-day performance ofthe TRAPATTdiode and three contemporary devices is illustrated infig. 1, for single diodes in pulsed oscillators. Althoughnot offering the very highest output powers, theT,RAPATT diode is more efficient than its. competi-tors, and can operate at higher duty cycles. Its opera-ting frequency, however, is limited to the lower end ofthe microwave spectrum. Despite the early promise ofthe device, and its attractions as a source for use in

[1] D. de Nobel and M. T. Vlaardingerbroek, IMPATT diodes,Philips tech. Rev. 32, 328-344, 1971.

[2] P. J. de Waard, Anomalous oscillations with an IMPATTdiode, Philips tech. Rev. 32:361-369, 1971.

[3] See for example B. C. DeLoach, Jr., and D. L. Scharfetter,Device physics of TRAPATT oscillators, IEEE Trans. ED-17,9-21, 1970.

100 R. DAVIES et al. Philips tech. Rev. 40, No. 4

pulsed radars, there has been no significant impactbeyond the results ofthe initial stages of development.Very few, if any, of the early TRAPATT oscillatorshave been manufactured in any quantity, and thereare few remaining TRAPATT studies.

This failure to make an impact on microwavesystems was due to difficulties in establishing andmaintaining coherent oscillations. (By 'coherent' ismeant a fixed-frequency oscillation in which the fun-damental and any harmonics have amplitude andphase relationships that are constant.) The lack ofprogress can be largely attributed to serious short-comings in the widely accepted design for the oscil-lator circuit. Circuits based on this design requiredempirical alignment, performance was unreliable, andcoherent operation into practical r.f. loads was diffi-cult to maintain. This article describes a new designthat overcomes these three disadvantages, providing afirm basis from which TRAPATT oscillators can befurther developed, to the point where routine produc-tion is possible.

In parallel with the progress in circuit design, diodedesigns have also advanced, to a stage where a rangeof silicon-planar devices can be mass-produced at lowcost. This progress has resulted from a coordinatedresearch programme that has highlighted critical fea-tures in a number of associated topics. We now havea fuller understanding of the underlying physical pro-cesses involved in the TRAPATT mechanism, leadingto a better appreciation of optimum 'geometry', ther-

104 WLSA

103

P

t 102

10

10-7

10~L-~ __~ ~ __~ __~ __~ __-L___WQ5 2 5 10 20 50 100GHz

-f

Fig. 1.Experimental results are summarized in this diagram as peakoutput power P plotted against frequency f for four kinds of micro-wave diode. These are the TRAPATT (Trapped Plasma AvalancheTriggered Transit) diode, the IMPATT (Impact Avalanche andTransit Time) diode, the LSA (Limited Space Charge Accumula-tion) diode, and the Gunn diode. Although the TRAPATT diodemay not appear to offer the highest peak powers in the band ofinterest (S-band, 2-4 GHz), it is more efficient than the Gunn diode,and can operate at higher duty cycles than the LSA device. Theseresults are for single devices. Comparable powers (+) can be ob-tained using bipolar transistors to amplify the output of a low-power microwave source, provided a combination of many tran-sistors is used.

mal design and choice of silicon material. Our studieshave led to a new diode structure that is more repro-ducible, more efficient, has lower thermal resistance,and together with the new circuit results in an oscil-lator that is considerably less temperature-sensitive.Present-day devices perform reliably at duty cycles ofup to 3OJo with heat-sink temperatures in excess oflOO°C, and can be produced in a range of chip sizes.The larger-area diodes and the new circuit have im-

pedances that closely match each other, and consider-ably higher r.f. powers can be obtained. Peak powerlevels of 300 W at 0.1% duty cycle and 250 W at 2%duty cycle can already be achieved at S-band (2-4 GHz)with single diodes, and increased power from inter-connected diodes has been demonstrated.

TRAPATT oscillator circuits

Basic principles

The TRAPATT oscillator is based on a reverse-biased silicon p-n junction that is operated at ava-lanche breakdown. This breakdown occurs when theelectric field-strength within the depletion region ofthe junction is sufficiently high that ionization pro-cesses give rise to a cumulative multiplication ofcharge carriers (holes and electrons). Fig. 2 shows thebasic TRAPATT diode, in which the n" substrate isbonded directly to the package, and the second con-nection is made via a gold wire, thermocompression-bonded to the upper contact. The active junction is atthe interface of the r" and n regions.The principles of circuit operation and the origin of

the device acronym can be understood most easily byconsidering the response of a p-n junction, biased justbelow avalanche breakdown, to an applied voltageimpulse. The impulse causes the electric field-strengthwithin the depletion region of the junction to rise

Fig. 2. The basic TRAPATT diode is manufactured using suitablydoped silicon and conventional processing techniques. It consists ofa p+n junction J lying on an n" substrate S. The package is con-nected directly to the substrate, and an ohmic contact C is providedfor a wire connection, through the surface oxide o. Typical verticaldimensions are shown in the figure (not drawn to scale). The diodearea is chosen to suit the particular application. This article dealsmainly with diodes of area 4x 10-4 cm>, and higher-power deviceswith areas of up to 12X 10-4 cmê.

Philips tech. Rev. 40, No. 4 TRAPATT OSCILLATOR 101

sufficiently for avalanche multiplication to occur. Asthe voltage applied increases, the avalanching regionspreads through the junction depletion region. Whenthis avalanching region expands faster than the maxi-mum velocity at which the carriers can be extracted(the saturated drift velocity), an extremely large num-ber of carriers is generated in the avalanche region,which is filled with a dense, electrically neutral con-centration of these charge carriers. -This is known asthe 'trapped plasma' state.When the associated high space-charge density be-

comes sufficiently large, the electric field within thedevice collapses to almost zero. Consequently the car-riers are extracted very slowly, at a rate much lowerthan that corresponding to the standard drift velocity.The electric field within the device gradually recoversas the carriers are extracted. To provide a suitablerecovery characteristic and to minimize resistive losses,

Q

~~_ 1-1C1L:L-t

Fig. 3. Diagrams to illustrate voltage variation with time t duringTRAPATT oscillation. a) The assumed train of positive voltagepulses Vp incident on the diode from the oscillator circuit. b) Thedevice voltage Vd, showing the transition to avalanche conditions inresponse to the voltage pulse, and the subsequent recovery. Theoscillator circuit is so designed that the negative-going pulses fromthe diode are inverted and reflected back towards the diode, thusforming the train of positive pulses assumed here.

I I

I I

-tT o

Fig. 4. Typical variation of the diode voltage Vagainst time t,during the establishment of coherent TRAPATT oscillations. Threephases can be distinguished: the initiation phase I in which anIMPATT oscillation possibly occurs, leading to a transient phase Tof non-coherent trapped-plasma oscillations, and the resultingTRAPATT oscillation 0 itself. Circuit designs should take accountof this three-stage behaviour.

TRAPATT diodes have an epitaxial-layer thicknessand doping such that the depletion region extendsthroughout the low-doped layer before avalanchebreakdown is reached [41.

Fig. 3 illustrates the response of the reverse-biaseddiode to an assumed train of positive voltage pulses,incident on the diode from the oscillator circuit. Thecircuit is designed to provide the necessary sequenceof voltage-triggering pulses, once the initial pulse hasbeen generated. This initial pulse originates within thedevice itself, possibly as the result of IMPATT oscil-lations. This has not been verified conclusively, al-though several suggestions have been made [51. Com-puter models have shown the TRAPATT oscillationdeveloping from a high-frequency IMPATT mode asindicated in fig. 4. to introduce the circuit concepts,let us assume the incidence of a trigger pulse, andsee how the circuit provides the conditions for self-sustained oscillations.The basic circuit is shown schematically infig. 5; it

is known as a Time Delay Triggered (or TDT) circuit.It consists of the TRAPATT diode coupled to a delayline and a lowpass filter. We have already seen howthe application of a voltage impulse causes the voltageacross the diode to collapse. This collapsing voltagecouples to the circuit delay line and a negative voltagepulse is launched along the line. This pulse travelsalong the line until it reaches the filter, which is de-signed to provide a resistive termination at the micro-wave output frequency and a low impedance at theharmonics of this frequency. (We shall see later thatenergy may be extracted at harmonics of the funda-mental frequency, but for the present we consideronly fundamental operation.) On reaching the filter,the negative pulse is inverted, partly reflected, andpartly transmitted to the load. The matching is ar-

1III

1'0 DL F

TI "'M2

Fig. 5. The three fundamental components of the Time Delay Trig-gered, or TOT, circuit are the TRAPATT diode D, a delay lineDL,and a lowpass filter F. The length of the delay line determines thefrequency of oscillation. Since each trigger pulse travels from thediode to the filter and back, the line should be of a length equal to

. half the wavelength at the desired frequency of oscillation. (In prac-tice it is slightly less than this, to allow for delays in the diodeitself.) A suitable externalload L is connected to the output of thefilter.

[4] A detailed description can be found in G. Gibbons, Avalanche-diode microwave oscillators, Clarendon Press, Oxford 1973.

[5] See for example B. B. van Iperen, Efficiency limitation bytransverse instability in Si IMPATT diodes, Proc. IEEE 62,284-285, 1974.

102 R. DAVIES et al. Philips tech. Rev. 40, No. 4

ranged to maximize the power dissipated in the loadat the fundamental frequency, while ensuring that thereturning positive pulse is large enough to trigger thediode directly into the TRApped Plasma state, andcause the Avalanche region to undergo a TriggeredTransit through the device. The subsequent collapseof the electric field within the diode generates a furthernegative pulse in the delay line and the process willcontinue until the d.c. bias is removed from the diode.This simple model shows that the TRAPATT oscil-lator is primarily circuit controlled and the frequencyof oscillation depends mainly on the electrical lengthof the delay line. This length is approximately onehalf of a wavelength at the fundamental TRAPATTfrequency.

The TRAPATT oscillations, once started, will con-tinue in the presence of the d.c. bias. The power avail-able is limited by the temperature rise in the diodestructure, even with an efficient heat sink. To limit thediode ternperature to a maximum of 200 oe at largebias currents, the oscillator is operated in a pulsedmode. In this, the bias is applied as a series of pulses;large peak powers can then be obtained. Pulse widthsof about 100 ns and duty cycles of a few per cent arecommonly used, corresponding to a pulse-repetitionfrequency of a few hundred kHz. Our investigationshave been confined to this mode of operation.

Circuit designApart from the diode, the only major components

of the TDT circuit in fig. 5 are the delay line and thefilter. It might appear that all that is necessary todesign a circuit is to choose the length and impedanceof the line, and the match and frequency pass band ofthe filter. A microstrip circuit designed according tothis simple concept would be as shown schematicallyin fig. 6. The length of the delay line would be deter-mined from the desired period of oscillation, and thefilter characteristic would be optimized experiment-ally. In practice, TRAP ATT oscillations were initiallyobserved in such a circuit, and the theoretical explana-

5~-~-----j -I ..

?./4

A/4BF

o DLlc==::::J -- LC

son

Fig. 6. This shows the basic microstrip oscillator layout, with a uni-form delay line DL and a lowpass filter F of unspecified geometry.The diode D receives its d.c. bias via the delay line, which is coupledto a d.c. source S via a bias filter BP. A blocking capacitor C isincluded at the output to prevent the bias from reaching the load L.

tion was developed only later. Until recently almostall the reported practical results were measured insuch circuits. Typical coaxial and micro strip versionsare shown in fig. 7. This circuit 'design' is based onthe following three criteria:

The circuit mustsupport the small-signal oscillations that grow totrigger the first cycle,provide enough locally stored charge to drive theplasma generation process,reflect the subsequent trigger pulses that maintainthe oscillation.

We find, using diodes with the correct dopingprofile, that the first criterion is satisfied in practicalcircuits that comply with the second and third criteria.We deduce this from the ease with which the devicevoltage collapses, which we interpret as indicatingTRAPATT action.

a

b

Fig. 7. Two versions of the original TDT oscillator. al The coaxialversion, and bl the microstrip version in which the delay line andfilter geometries can be dearly seen. The delay line has a 30nsection adjacent to the diode, which was preferred to direct con-nection to a son line. The resulting 30Q to son step just to the leftof the first large low-impedance section was found to be effective inpreventing the diode from breaking down during the recoveryperiod.

The second criterion is confirmed by the improve-ment in oscillator performance that results fromreducing the delay-line impedance near the diode.

Satisfaction of the third criterion has been widelyinterpreted by TRAPATT circuit designers as requiringtwo circuit elements. The first is a delay line that isslightly shorter (by about 10070 at S-band) than a half-wavelength at the TRAPATT frequency, to define theperiod of the trigger pulse. This shortening is to allowfor the delay incurred by the device during the chargingand avalanche-multiplication period. The second ele-ment is a filter terminating the delay line. This pro-vides a resistive termination at the output frequency(which is usually the fundamental TRAPATT fre-quency but might be at a higher harmonic frequency),and a low impedance at other frequencies harmonic-ally related to the fundamental.

There are two major shortcomings of this circuit.One is that when using a broadband-match load,

Philips tech. Rev. 40, No. 4 TRAPATT OSCILLATOR 103

adjustment for coherent oscillation is completelyempirical. The other is that subsequent operation intopractical loads (e.g. isolators, bandpass filters etc.)is usually incoherent, necessitating a further adjust-ment that often degrades the efficiency. Before theTRAPATT oscillator could become more than a'laboratory curiosity', a circuit design that gave pre-dictable and reproducible interaction with the diodewas essential. Such a design has resulted from time-domain studies we have made of TRAPATT circuitsand associated components.If practical loads such as filters and isolators are

measured in the-time domain with, for instance, areflectometer, it is found that they have broadlysimilar characteristics. These loads can present areactive mismatch to the oscillator lowpass filter,giving rise to spurious trigger pulses. In such cases,the diode may receive two or more trigger pulses percycle, one wanted and the others unwanted. Therewill be no adverse effect if the wanted pulse dominates,or if the pulses are nearly coincident. However, theusual result is that coherent oscillations cannot beestablished.It is the low-frequency energy content of the output

signal during the transient phase that gives rise to thisproblem, and once this is realized, the cure becomessimple. A diplexer, connected between the oscillatorand load, is used to divert the low-frequency signal toa second, matched, load. A suitable component forperforming this diplexing function is the travelling-wave directional filter (TWDF) [61. This is a four-port component consisting of two transmission linescoupled via a resonant ring; although a resonantdevice, it has a constant broadband input resistance.Use of this component gives complete suppression ofthe spurious pulses, and has allowed operation intopractical loads such as isolators and filters. A useful'spin-off' is that the d.c. bias for the TRAPATT diodecan also be fed through the TWDF, in such a way thatnone of the r.f. output power is dissipated in the biascircuit.A 30n to son step was usually built into the delay

line between diode and filter, as in practice this madeit easier to establish coherent oscillations.

In the light of these results the design criteria havebeen modified to read as follows:The circuit must

- support the small-signal oscillations that grow totrigger the first cycle,

- generate the optimum voltage-time response inboth the transient and steady-state phases,

- reflect the subsequent trigger pulses that maintainthe oscillation, without introducing spurious trig-ger pulses.

a.

p; -------,.810Q .-----~50n _ ____l

=w b

Fig. 8. Diagrams illustrating the construction of the new oscillatorin its two forms: a) shows a cross-section of the coaxial version, inwhich a packaged diode D is connected to one end of a coaxialdelay line comprising a constant-diameter inner conductor I en-closed in a stepped outer conductor O. A IOn section is includedjust before the son connector C. b) shows the microstrip version,in which a diode chip D is connected to one end of a linearly-tapered delay line DL. A IOn section is included before the sonline, which leads to a coaxial connector (not shown). Both versionswould have their output connected to the load via a travelling-wavedirectional filter. .

The simple circuits shown schematically in fig. 8have resulted from the application of these newcriteria. They differ from the conventional TRAPATToscillator in two significant aspects: the delay line isstepped or tapered to prevent the diode voltage fromrecovering too rapidly, and there is only one elementin the output filter so that spurious trigger pulsesare eliminated from the circuit. This design there-fore takes into account the way in which coherentTRAPATT oscillations grow during the transientphase, and produces a circuit that does not sufferfrom the disadvantages inherent in the earlier designs.

Oscillator construction and performance

Microstrip and coaxial versions of the new design ofoscillator are shown in fig. 9. Whereas the microstripversion incorporates a tapered delay line, the coaxialoscillator has a delay line with five discrete impedancelevels. The basic electrical characteristics of theseoscillators are, however, identical, and are summarizedin Table I for oscillators using single diodes of area

. 4 X 10-4 cm",The inherently lower losses in the coaxial transmis-

sion-line allow larger area devices to be used, witha consequent increase in output power, and so moreattention has been paid to this version than to themicrostrip one. The coaxial version is also more suit-able for testing performance reproducibility since itcontains packaged devices.

[6) s. B. Cohn and F. S. Coale, Directional channel-separationfilters, Proc. IRE 44, 1018:1024, 1956.

104 R. DAV lES et al. Philips tech. Rev. 40, No. 4

The reproducibility of the power and frequencycharacteristics has been measured using a fixed-tunedversion of such an oscillator. The measurements weremade using diodes that were inverted so that their p+

Fig. 9. Practical embodiments of the new oscillator design. Thetapered delay line used in the microstrip version M is clearly visible.The enclosed coaxial version C incorporates a similar non-uniformline having five distinct steps. The basic electrical characteristics ofthese two designs are identical.

Table I. Basic electrical characteristics of oscillators made to thenew design, using diodes of area 4 x 10-4 cm2. The peak powerincludes the losses in the TWDF, and is measured at the output ofan isolator which is connected to the output of the oscillator duringthe measurement.

Frequency:

Peak power:

Efficiency:

Duty cycle:

Pulse length:

df/dT:

Temperature range:

S-band

60 W minimum

30070 typical

2070 maximum

300 ns maximum

-450 kHz °C-1 maximum

- 54°C to + 95 oe

Fig. 10. The dependence of peak power P and efficiency IJ on a) the resistive load R, and on b) thedrive current I, also showing the frequency variation I'lf. In both cases, further increase of theindependent variable caused loss of coherence.

100W

p

t 80

9

100W

20MHz

p L1f

t10

t800

40% tO%

Ti -10~~'l ~30t

60~_x--' ry - 30 t-20

20 20

10 40

2 3 4 5 6Q 35 4.0 4.5 SOA-R Q -1

60

40

contact (see fig. 2) could be bonded directly to theheat sink to minimize the thermal resistance. Fourteensuch 'flip-chip' diodes of area 4 x 10-4 ern", and fromthe same batch, were inserted sequentially. For eachdevice, power and frequency were measured for a4 A, 0.1 % duty cycle. An acceptably small spread of± 5% in power was measured; the frequency spreadwas ± 23 MHz.

Using the coaxial-oscillator design, it has been pos-sible to study how the oscillator characteristics relateto the circuit load at the fundamental and harmonicfrequencies, and to investigate the tolerance tochanges in drive current. The resistive loading at thefundamental frequency was varied by adjusting thelength of the low-impedance step and the configu-ration of the delay line. A current drive of 4 A wasused at a duty cycle of 0.1 %. Fig. la shows how thepeak output power and the efficiency increase withincreasing resistive loading. It was not possible tomaintain coherent operation for resistive loads inexcess of 5.3 Q, because premature avalanching wasbrought about by the excessively large output voltageacross the diode. It was also found that coherent oscil-lations were not critically dependent on the loading atthe harmonic frequencies. The tolerance of the oscil-lator to variation in the drive conditions is also givenin fig. 10, which shows how the power output, effi-ciency, and frequency vary with increasing currentdrive for a constant circuit configuration. The poweroutput and efficiency increase until coherence is de-stroyed by premature avalanching at a peak currentslightly in excess of 5 A.

Philips tech. Rev. 40, No. 4 TRAPATT OSCILLATOR 105

Oscillator tuning

It is necessary to provide a simple means of tuningan oscillator if it is to be used in a practical systemoperating at a specified frequency. Oscillators arecommonly tuned by varying the fundamental reso-nance of the frequency-determining components. TheTRAPATT oscillator is. extremely rich in harmonics,and it can be tuned by modifying either the funda-mental resonance or the resonance of one or more ofthe harmonics, or by a combination of these methods.

The frequency of a TRAPATT oscillator is deter-mined by the delay introduced by the circuit and thedevice. To tune the oscillator, the circuit delay can beadjusted by varying the electrical length of the delayline, or the device delay can be varied via the har-monic terminations.

It is worth noting that electronic tuning by meansof a varactor diode (which has a voltage-dependentcapacitance) is not particularly suited to a TRAPATToscillator. This is because the varactor diode is essen-tially a small-signal tuning element, and the high peakpower available from TRAPATT oscillators limits themethod to very small tuning ranges.

Mechanical tuning

Mechanical tuning is achieved most readily byadjusting the termination of as many harmonics aspossible. In the TDT circuit this is accomplished byadjusting the length of the delay line via the positionof the low-impedance step. A coaxial oscillator thatcan be tuned in this way is shown in fig. vu. The figurealso shows the expected linear relationship betweenfrequency and delay-line length, with an accompany-ing power variation of about 10 W over the tuningrange.

Magnetic tuning

There are two methods of tuning in which an ex-ternal magnetic field is varied. The required fieldchange in each method is small, and could be obtainedfrom a small field coil. The tuning rate is reasonablyhigh, and may reach several tens of megahertz permicrosecond. Both methods of tuning have beeninvestigated, using the same basic TDT oscillator.

The first method is tuning by variation of effective.permeability, and was originally investigated for thetuning of IMPATT oscillators [7] and later forTRAPATT oscillators [8]. In our experiments we useda micro strip oscillator constructed on a ferrite sub-strate [9]. The electrical length of the delay line is afunction of the permeability of the substrate material,and it follows that the line length may be varied by theapplication of an external magnetic field. The con-struction of the prototype oscillator is shown in fig. 12.

40 rwp20 50 t0 4054 55 56 57 58 59 60mm

-L

Fig. 11. The photograph shows a mechanically-tuned coaxial oscil-lator in which the position of the low-impedance step can bemanually adjusted by means of the knurled section S. A linearvariation of frequency àf of over 100 MHz is obtained when thedelay-line length L is changed by about 4.5 mm. There is a reduc-tion in output power P as the frequency rises, but this is acceptablysmall.

~p

tH

~---~====~I

I

~p

Fig. 12. The TRAPATT oscillator on its ferrite substrate FS islocated between the poles P of an electromagnet. The oscillator istuned by varying the applied magnetic field H.

[7] B. Glance, A magnetically tunable microstrip IMPATT oscil-lator, IEEE Trans. MTT-21, 425-426, 1973.

[8] S.-G. Liu, Magnetically tunable TRAPATT oscillator, 1974IEEE Int. Solid-State Circuits Conf. Dig. tech. Papers, pp.98-99.

[9] P. L. Booth, S. R. Longley and B. H. Newton, Frequencytuning of microstrip TRAPATT oscillators, IEEE Trans.MTT-29, 6-10, 1981.

106 R. DAV lES el al. Philips tech. Rev. 40, No. 4

Varying the magnetic field by 60 kA m-I gave theresults shown infig. 13. The parameter on the graph isthe angle e between the normal to the magnetic fielddirection and the plane of the substrate. Different set-tings of e were investigated to see if an optimum valuecould be found. The widest tuning range was 61 MHzfor e = 18°, starting at a frequency of 2.19 GHz. Theoutput power at this setting remained substantiallyflat (13 W peak ± 0.5 dB) over the frequency rangeand the frequency spectrum was clean and symmet-rical. The tuning curves show a pronounced 'dip' andare very nonlinear. We have found that a qualitativeexplanation of the shape of the curves is possible if thedemagnetization effect of the substrate is considered.

f

r 220

B /Jf (MHz)

0° 1.8gO 1.618° 6190° 1.1

a20 40 60kAm'

-H

p

t

20 40 6OkAmo,-Hb

c 10fvlHz -f

Fig. 13. Experimental results for magnetic tuning by varying theeffective permeability. The angle e between the ferrite substrate FSand the normal to the direction of the magnetic field H was set suc-cessively to 00,90,180 and 90°. a) The optimum setting of 18° gavethe largest change of frequency f. b) This setting also correspondedto the smallest variation in peak output power P. The oscillator wasadjusted for good spectral purity rather than for maximum outputpower. c) The power-output spectrum is clean and symmetrical, ascan be been on this photograph of a measured relative power P,ver_sus frequency f.

The second magnetic-tuning method makes use ofthe harmonic terminations. By computer modelling, ithas been shown [10][11] that frequency tuning is pos-sible by varying the amplitude or phase of the funda-mental, 2nd, 3rd and 4th harmonics. This simulationpredicted that tuning by phase variation at the 3rdharmonic gives the largest tuning range with the leastvariation in output power. Altering the amplitude ofthe 3rd harmonic causes a smaller change in fre-quency.

Magnetic tuning via the harmonic terminations canbe achieved by using polycrystalline yttrium iron gar-net (YIG) spheres as a ferrimagnetic resonator, andthis is the method we have adopted (12]. The practicaloscillator had two YIG spheres of 0.93 mm diameter.They were semi-loop coupled approximately onequarter of a wavelength from the open-circuit end ofan air-spaced micro strip transmission line connectedin parallel with the TRAPATT diode chip. The diodewas also connected in parallel with the standardmicrostrip oscillator circuit. This is shown in fig. 14.The magnetic field required to tune the ferrimagneticresonator, which primarily controls the oscillator fre-quency, was applied perpendicular to the substrateand parallel to the plane containing the semi-loops.The layout of the oscillator was adjusted to give thebroadest possible tuning range.

Using this method, we obtained a fundamental-fre-quency tuning range of over 120 MHz at a centre fre-quency of 2.46 GHz, by biasing the spheres to resonatenear to the third harmonic. The tuning curves forpower and frequency are illustrated infig. 15. It is alsopossible to use a simpler arrangement with a singlesphere, but this has a smaller tuning range of about90 MHz, and produces about 10 W less output power.The d.c. to r.f. conversion efficiency of these circuitsis typically between 13 070 and 23%.

Fig. 14. This figure shows the positioning of the two YIG spheres S,with their semi-loops SLo The quartz plate P provides a mechanicalsupport for the air-spaced transmission line T, The TRAPATTdiode D lies at the end of the delay line DL, on an alumina sub-strate AS.

Philips tech. Rev. 40, No. 4 TRAPAIT OSCILLATOR 107

2.54GHz

2.42 ",/+_...-""T

f

t 2.50 48W

44 p

40 t36

2.4.6-c-- .......---~

2.380'------'-5--10'---1-'-5-----'2'-0-___J25 k/vm"

-.£lH

Fig. IS. Variation of frequency f and peak output power P withchange in magnetic field I1H. The nearly linear relationship betweenfrequency and field results from the fact that the sphere resonancefrequency also increases linearly with the field.

Compared with the variable-permeability approach,this novel method provides somewhat larger tuningranges with improved linearity, and demonstrates thefeasibility of individually tuning the harmonics of aTRAPATT oscillator.

Frequency-stabilized TRAPATT oscillators

The frequency of a TRAPATT oscillator is a func-tion of circuit and device delays, each of which is tem-perature dependent. In practice the variation' withtemperature of the circuit delay is negligible. Thedevice temperature dependence leads to an increaseddelay with increasing temperature, which results in anegative variation of frequency with temperature,dJ/ dT, of up to 450 kHz °C-l at S-band. This fre-quency variation manifests itself in two distinct ways.There is a variation or 'chirp' during the r.f, pulse,due to heating of the device by the passage of current,and there is a variation with ambient temperature.Three circuit techniques have been studied forreducing the oscillator frequency dependence on tem-perature: injection locking, temperature compensa-tion and temperature stabilization. Of these, onlyinjection locking affects the characteristics of thechirp.

Injection-locked oscillators

Typically, at an operating current density of10 kAcm-2, the chirp during the 0.5 IlS bias of a free-running 2.47 GHz oscillator was 6 MHz. For mostradar transmitters, however, a frequency stability of1 MHz or better is required. Injection locking is atechnique whereby a free-running high-power oscilla-tor is coupled to a stable, low-power oscillator in such

a way that a stable, high-power oscillation is pro-duced. Fig. 16 shows a typical 'locking' characteristicfor a microstrip TRAPATT oscillator having a 0.5 IlSbias pulse and a pulsed locking source. An almostlinear relationship exists between the logarithm of thelocking range and the locking gain from 15 to 20 dB.At higher gains the relationship becomes nonlinearowing to the inherent frequency chirp. A system hasbeen devised in which the locking source is itselfphase-controlled using a discriminator at 1.3 GHzand a crystal reference [13]; Such a system can have afrequency stabilility of ± 5 ppm or better.

100MHz~---------~

R

t10

10'-----'----'----'--~10 20 30dB

-G

Fig. 16. The locking characteristic is the usual way of representingthe behaviour of a free-running oscillator that is coupled to a low-power locking oscillator via a non-reciprocal device such as a cir-culator. Such circuits produce a stable, high-power oscillation. Thelocking range R is that range of locking-oscillator frequencies forwhich the free-running oscillator becomes synchronized. Thelocking gain G is the oscillator power ratio (free-running/locking).A simple theory predicts that a linear relationship exists betweenthese two parameters, the locking range reducing by 1 decade forevery 20 dB increase in locking gain. This is shown by the dashedline in the figure. The solid line shows that the measured per-formance of a TRAPAIT oscillator is in good agreement with thistheory. (The measurements apply to a 2.47GHz microstrip oscil-lator with a peak power output of 13.3 W, and operating at a dutycycle of 0.1 0/0.)

Temperature-compensated oscillators

The change of frequency of a TRAPATT oscillatorwith temperature can be compensated, by mechanic-ally retuning the oscillator in sympathy with changesin ambient temperature. This is most easily achievedby arranging for the position of the low-impedance

[lOJ R. J. Trew, G. I. Haddad and N. A. Masnari, The operationof S-band TRAPAIT oscillators with tuning at.multiple har-monie frequencies, IEEE Trans. MTT-23, 1043-1047, 1975.

[liJ R. J. Trew, Properties of S-band TRAPAIT diode oscillators,thesis, University of Michigan 1975.

[12J S. R. LongleyandP. L. Booth, FrequencytuningofTRAPAIToscillators using ferrimagnetic resonators, 8th Eur. MicrowaveConf., Paris 1978, pp. 790-794.

[lSJ B. H. Newton and G. Payne, A rugged phase-locked C-bandsource, Mullard Research Labs Annual Review 1973, pp.88-94.

108 R. DAVIES et al. Philips tech. Rev. 40, No. 4

step to be temperature-dependent. The thermal coeffi-cient of expansion of different materials can be ex-ploited for this; plastic and metal are used in the com-pensated oscillator shown in fig. 17. This oscillatorchanged by only 4 MHz in frequency as the ambienttemperature was varied from 20 oe to 100 oe, com-pared with 30 MHz for the uncompensated oscilla-tor [*1.

Fig. 17. The thermal expansion of different materials can be used tocornpensate for the change in frequency of this oscillator as thetemperature varies. The rods R that surround the coaxialline L aredivided into two groups of seven each, the groups acting in parallel.In each group, the rods are connected serially (end-to-end), in a zig-zag fashion to save space. One extremity of the group is fixed to thebody of the oscillator, and the other to a moveable low-impedancestep. By using alternate rods of lnvar and a chosen material, anadditive expansion is obtained that can be made to match the oscil-lator characteristics very closely. In this example, the materials usedare Delrin and aluminium.

Temperature-stabilized oscillators

The temperature of the diode itself can be stabilizedat an elevated value, by using the heating effect of asmall reverse current that is allowed to flow throughthe diode. An external control circuit is employed thatdecreases the current as the ambient temperature rises.It is sufficient to maintain a simple linear relationshipbetween this current and the temperature. Because thecurrent is applied between oscillator pulses, and is atleast an order of magnitude smaller than the thresholdcurrent for TRAPATT oscillation, no spurious r.f.power is generated. This technique gives a stabilizedfrequency over a wide range of ambient temperatures.

Device manufacture

The initial theoretical and experimental work de-monstrated the potentialof the TRAPATT diode forthe efficient generation of high peak power at micro-wave frequencies. However, it became clear that toproduce acceptable devices, careful design was neededto accommodate the large electric field and high cur-rent density, both inherent to the TRAPATT mechan-ism. So in addition to investigating the electrical de-sign, it has been necessary to consider reliability, per-formance reproducibility and high-yield processing.

The result is that devices have been available for sometime for circuit investigations, and are now beingdeveloped for specific systems applications.

Diode structures

Unless special precautions are taken, the electricfield-strength at the edge of an avalanche diode can behigher than elsewhere, leading to localized prematurebreakdown ('edge breakdown'), which destroys thedevice. It is therefore of prime importance to designthe diode so that the field is low at the edges, anduniform over the junction region. Two constructionshave been investigated that can be made to satisfy thisdesign requirement. These are the mesa and deep-diffused planar constructions.

The SEM (Scanning Electron Microscope) picturein fig. 18 shows a typical mesa diode. The edge of the

Fig. 18. An SEM picture showing a mesa diode profiled to preventedge breakdown of the junction. The insert shows a positive bevel(a) that ensures that, in the active part of the device, the p-njunction has the largest area. This is not the case for a negativebevel, (b). The photograph corresponds to stage C in fig. 19, with-out passivation.

diode is bevelled positively so that the p-n junction isalways the largest area in the diode. For maximumelectrical efficiency, the bevel angle is critical and thesurface of the mesa must be smooth. If the mesa iswet-etched these conditions can be difficult to controlbecause of electrochemical action in the vicinity of thep-n junction. In extreme cases this can produce alocalized negative bevel. An investigation of etchingeffects using a dry gas plasma has shown that the bevelangle can be controlled accurately depending on thecomposition of the etching gas. This has made pos-sible a new approach to mesa technology, in which thediode can be etched from the junction face as opposedto conventional etching from the substrate. Using theprocesses shown infig. J9, mesas can be accurately de-fined from the junction face by whole-slice processing

Philips tech. Rev. 40, No. 4 TRAPATT OSCILLATOR 109

g

J n ln+

3~+ (

d

Fig. 19. Processing steps used in producing a plasma-etched mesadiode: a) n" substrate with n-type epitaxial layer, and maskinglayers Si02, Si3N4, Si02. Photolithography is used to define themesa mask. b) Mesa definition by plasma etching. c) Mesa shapingby selective plasma etching to give the mesa the required outline.Si02 passivation is then applied. d) p-region formed either bydiffusion or ion implantation. Metallization M applied for ohmiccontacts, plus a plated gold layer G. e) Device inverted on to mainheat sink H.

either before or after heat-sink attachment. Theresulting devices are completely free of any surfacecontamination produced during the etching process.An advantage of plasma etching is that junction de-sign and edge-breakdown criteria can be optimizedindependently during processing.

In 1968, a method was described [14) for makingplanar IMPATT diodes with improved breakdowncharacteristics. This was achieved by diffusing thejunction through an oxide window to a depth suffi-cient to prevent edge breakdown. The same methodcan be used to make planar TRAPATT diodes. Inpractice, the junction depth is determined by thedoping of the n-type silicon and so this requirementand the junction design cannot be independently op-timized. The end regions of the diode are thereforecharacterized by graded profiles, and the depth of theactive junction region makes it difficult to provide thediode with an effective heat sink. However, the diodecan be made by conventional silicon-planar proces-sing and has been shown to be ideal for high-yieldbatch processing. Provided long-life metallizationschemes are incorporated, this results in reproducibleand reliable devices.

A simple, but effective, flip-chip device has beenevolved using planar processes for all the fabricationincluding the heat sinks. This device is well-suited forapplications with duty cycles of less than 2!IJo, andtypical performance figures are shown in Table I.Thelimiting factor in the structure is the thick diffusedp-region that lies between the n-region, where the heatis generated, and the gold heat sink. This gives a mini-mum chip thermal resistance of 10 oew-1 for a diodearea of 4 x 10-4 ern". Additionally, the gold heat sinkdegrades the performance of the diode by reducingthe d.c. to r.f. conversion efficiency by about 5!IJo.This occurs because, in the region where the heat sinkoverlays the oxide, there is a significant oxide capac-itance in parallel with the junction, and this capac-itance increases the charging time of the diode duringthe charge-generation period. Thus the design alwaysinvolves a compromise between thermal spreadingresistance and electrical efficiency.

To overcome the electrical and thermal limitationsof the flip-chip diode, a more ideal' thinned' structurehas been produced. This is necessary for operation athigher duty cycles. The new device, which is shown infig. 20, has a lower thermal resistance (5 oe w-1 for adiode of area 4 x 10-4 cm") but the electrical efficiency

nj en]? ,~i;mp+ n

IH

1f.lm

C yGa

c

H bD

Fig. 20. The 'thinned' TRAPATT diode. a) A schematic cross-section showing the separation of only 1 urn between the junctionregion and the silver heat sink H. Ohmic contacts C are provided oneach side of the diode, and the heat sink is gold plated (C) aboveand below. b) An SEM picture of such a diode D, showing thesilver heat sink H and the upper ohmic contact C.

[*1 The temperature-compensated oscillator was designed andmeasured by G. Tubridy.H. G. Kock, D. de Nobel, M. T. Vlaardingerbroek and P. 1.de Waard, Continuous-wave planar avalanche diode withrestricted depletion layer, Proc. IEEE 56, 105, 1968.

(141

110 R. DAVIES et al. Philips tech. Rev. 40, No. 4

is not degraded by the heat-sink arrangement. Thedevice is identical to the planar diode, except that allbut 1 urn of the n" substrate is removed, and the re-:maining layer is connected directly to the heat sink.The distance between the active junction and the heatsink is thus reduced from 10 urn to 1 urn, arid theassociated decrease in thermal resistance has beenmeasured to be greater than 500/0. As the main heatsink is connected directly to the n" substrate, the strayheat-sink capacitance is completely eliminated, givingan improvement of at least 5% in the electrical effi-ciency of the diode. A further advantage is thatremoval of the bulk of the substrate reduces the seriesresistance to less than 1Q, and so the circuit loading isimproved. It is then possible to achieve impedancematching to larger area diodes, and so higher outputpower can be obtained than for other designs. Thethinned diode can be designed so that its oscillatingfrequency is less temperature-dependent than that fora conventional device. A dJ/ dT of only 100 kHz °C-1

is possible, compared with 450 kHz °C-1 for flip-chipdevices. Some thinned devices have shown almostzero dJ/ dT, and work is in progress to improve thediode design until this valuable characteristic becomesstandard.To produce thinned devices, it has been necessary

to develop a technology in which whole silicon slicesare reduced to a uniform thickness of 15 urn ± 1 urnbefore the heat-sink plating is formed. A high-yieldprocess has been established, and devices of this typeare in development at Mullard Hazel Grove.The performance and likely applications of flip-

chip and thinned diodes are given in Table 11.

Table 11. A comparison between the performance and likely appli-cations of flip-chip and thinned diodes.

Flip-chip Thinned device

Output power: >60W (2-3 GHz)Duty cycle: 2% maximum

>60 W (2-3 GHz)>20/0

>35%Suitable for applications re-quiring longer duty cycles,such as airborne radars. Alsocapable of further develop-ment for phased arrays andmarine radars

Efficiency:Applications:

35% maximumTypicallylow costradars, such asaltimeters, port-able radars, etc.

High-power' oscillators

There are basically two ways of increasing theoutput power of a. TRAPATI oscillator, either bychanges to the diod~structure, or by using a combina-tion of diodes. Of course, both these approaches can.also be used at the same time. Changes to the diode

structure include increasing the area of the device andimproving the d.c. to r.f. conversion efficiency.Diodescan also be combined in a number of ways, and this ispossible either at chip level(an interconnected arrange-ment of chips within one encapsulation), or at circuitlevel (an interconnected arrangement of packageddevices).

An important quantity underlying this subject is theimpedance of the device itself. For efficient transfer ofpower, a match must be maintained between thediode or diode combination, and the rest of the oscil-lator circuit. There is a finite minimum circuit impe-dance that imposes a limitation on some of these ap-proaches, despite advances in oscillator circuit design.A diode with an intrinsically higher impedance there-fore has a clear advantage.

Improved devices

Peak output power is roughly proportional to devicearea, the practical limitation to power being deter-mined by the lowest impedance that can be matchedby the microwave circuit.

All the results that have been discussed so far arebased on diodes with an area of 4 x 10-4 cm". We haveinvestigated larger devices; in particular, deviceshaving no heat sink, and with junction areas of upto 12 X 10-4 cm", have been measured in an S-bandcoaxial circuit. The oscillator was optimized by ad-justing the length of the low-impedance step to maxi-mize the output power. The power, efficiency andasssociated drive current are shown in Table III [**1.These results all correspond to the same operating fre-quency, and since the efficiency is maintained in thelarge-area devices it would appear that appreciablylarger devices could be used before circuit lossesbecome significant.

Optimization of the charge generation period ofthe diode would increase efficiency. With the presentdiodes there is a compromise in the design betweencharge generation and charge removal. Refinement ofthe carrier avalanche processes gives an improvementin the dynamics of plasma formation, and increasesthe density of the trapped plasma. These effects can bedemonstrated by optical illumination of the diode,and one approach may be to integrate a solid-state

TableHl, Experimental results for a range of diodes with differentjunction areas.

Diode area Peakpower Efficiency Current(cmê) (W) (0J0) (A)

4x 10-4 150 22 108 x 10-4 240 25 14

12 x 10-4 320 28 16

Philips tech. Rev. 40, No. 4 TRAPATT OSCILLATOR III

laser for independent control and modulation of thecharge generation.

Combinations of devices

Various circuit and device options exist for paral-lel/series combinations that give more power than asingle device. We have investigated several intercon-nections of two or four small-area devices, at frequen-cies in the range 2.1 GHz to 2.4 GHz. The circuitsused have been in both coaxial and micro strip con-figurations, and to make power optimization andalignment easier the circuits had mechanical tuningsections as previously described. All our measure-ments have been made at room temperature, usingbroadband load resistances, with duty cycles in therangeO.05OJo to 1%.

Arrangements of two 2x 10-4 cm'' diodes connectedin parallel give peak output powers between 50 Wand70 W, with an efficiency of about 23%. The construe-

Fig. 21. Two TRAPATT diode chips Dl and Dz connected in series.The gold-plated type llA (high thermal conductivity) diamond heatsinks Hare thermocompression bonded to the mount M and thechips are flipped and bonded to the upper surfaces. (Note that theleft-hand diamond is metallized on all six surfaces, to provide elec-trical connection to ground.) Connection to the circuit is madefrom the back contact of the right-hand chip, but is omitted herefor clarity.

Fig. 22. An SEM picture of a series-parallel assembly of 4 diodes (inpairs D1-Dz, D3-D4) interleaved with gold-plated metal discs M.This combination has the sarne impedance as that of a single diode,but must operate at low duty cycles to avoid excessive heating of theupper diode pair.

[u] P. L. Booth and G. Tubridy supplied the results for the high-power devices.

tion of a series-chip connection in the micro strip cir-cuit is shown in the SEM picture of jig. 21. The diodesare each of area 4 x 10-4 ern", and this arrangementproduced a peak output power of 100 W at a drivecurrent of 4 A, with an efficiency of 20%.

Another approach, in which four diodes present animpedance of just one device, is shown in jig. 22.Clearly this series/ parallel stacked arrangement doesnot give the ultimate in thermal resistance but the useof diamond offers the possibility of combining severaldevices and yet retaining a practical impedance level.These diodes combine to give four times the powerfrom a single diode, i.e. a total of between 200 Wand 240 W. Because the two uppermost chips have noheat sink, operation was possible only at low dutycycles between 0.05 % and 0.1 %. The efficiency wasabout 23%.

For comparison, single diodes from the same batchwere operated in standard coaxial and microstrip cir-cuits, giving typically the following power outputs at4A drive current:

coaxial microstriparea: 2 x 10-4 cm''area: 4 x 10-4 cm2

40W70 W

30 W60W

The results show that the output power can bereadily increased by using combining techniques atchip and circuit level. Series, parallel and series-paral-lel chip combinations are possible. It can be concludedthat several hundred watts of peak power with meanpower levels of the order of several watts at S-bandare attainable, using these techniques with currentlyavailable TRAPATT diodes. The mean power capa-bility of the diamond heat-sink configuration is cur-rently under investigation.

Much of the work described in this article has beencarried out with the support of Procurement Executive,Ministry of Defence, sponsored by DCVD.

Summary. The TRAPATT oscillator is a source of high peakpowers at microwave frequencies in S-band (2-4 GHz). It is basedon the TRAPATT diode, and is particularly suitable for pulsedapplications in radar systems. Although operation is at lowermicrowave frequencies than some contemporary oscillators, effi-ciencies are high and comparatively large duty cycles are possible.The conventional TRAPATT oscillator circuit has shortcomingsthat make its use in practical systems intractable; in particular , lossof frequency coherence arising from spurious trigger pulses is acommon problem. A new circuit has been developed that obviatesthese difficulties, and which has a much improved performance.Coaxial and microstrip versions of the oscillator have been made tothis new design, and various aspects of their performance have beenmeasured. The oscillator can be tuned either mechanically or mag-netically, and can be made insensitive to ambient temperaturevariation in three different ways. Considerable attention has beenpaid to the design of diodes with optimum electrical and thermalcharacteristics; this has resulted in a new 'thinned' diode structure.Very high peak powers can be obtained using both large-area devicesand novel circuit configurations.