1 MAPLD 2004 - C192Degalahal

SESEE: A Soft Error Simulation & Estimation Engine

V Degalahal1, S M Çetiner 2, F Alim 2, N Vijaykrishnan1, K Ünlü 2, M J Irwin1

1Emerging and Mobile Computing Center(EMC^2)

2 Radiation Science and Engineering Center (RSEC)

Pennsylvania State University

MAPLD 2004

Outline

Introduction Soft Errors: Physics

Scattering Absorption

Monte Carlo Method: Overview Tool Outline

Neutron Transport Simulation Modeling Neutron-Si interactions Ion transport simulation Circuits simulation

Case Study

Introduction

Soft errors or transient errors are circuit errors caused due to excess charge carriers induced primarily by external radiations

These errors cause an upset event but the circuit itself is not damaged.

Same as SEU (single event upset) SEU for space-born errors

Soft Errors

For a soft error to occur at a specific node in a circuit, the Qcollected at that particular node should be more then Qcritical

As CMOS device sizes decrease, the charge stored at each node decreases (due to lower nodal capacitance and lower supply voltages).

This potentially leads to a much higher rate of soft errors

Soft Errors

Soft Errors can cause problems in 3 different ways Affects memory elements like caches and

memory Affects the data path if the error propagates to

the pipeline registers. Change the character of a SRAM-Based FPGA

circuit(Firm Error)

B

S D

p substrate

G

n+n+

n channel

Soft Errors

+ - + -+ -

+ -+-

+ -+ -

+ -+ -

A particle strikeCurrent

3.6 eV for one electron hole pair

Interaction Mechanisms

Scattering Elastic scattering

Potential scattering Resonance scattering Interference scattering

Inelastic scattering

Absorption

Elastic Scattering

Potential Scattering Neutron scatters elastically off of the nuclear

potential without penetrating the nuclear surface

Resonance and Interference Scattering The neutron is first absorbed by the target

nucleus creating a compound nucleus Creation of compound nucleus is then followed

by the reemission of neutron The target nucleus returns to its ground state.

Energy Transfer throughInelastic Scattering

The incident neutron is first absorbed by the nucleus forming a compound nucleus

The nucleus subsequently decays by reemitting a neutron

Unlike elastic scattering, the final nucleus is left at an excited state

nucleon target X1X2 Xn residual nucleus

Energy Transfer throughInelastic Scattering

Such reactions occur only for relatively high energies.

Kinetic energy is not conserved When the incident energy of the neutron

exceeds 280 MeV, secondary pions can also be produced

The identity of the incoming particle is lost, and the creation of secondary particles is associated with energy exchanges of the order of MeV or larger

Energy Transfer through Absorption

Incoming particle is captured by the nucleus. The absorption might be followed by a subsequent

gamma emission depending upon the state of the compound nucleus

The absorption process of our interest is the 10B fission capture:

keV) (1776.73 keV) (1013

keV) (478 keV) (840 keV), (1472

47101

7*74*7101

HeLiBn

LiLiHeLiBn

Monte Carlo Method

Utilizes Stochastic Technique based on the use of random numbers and

probability statistics to investigate problems Obtains answers by simulating individual

particles and recording some aspects of their average behavior

Does not solve an explicit equation

Monte Carlo Method

Does not need averaging approximations required in space, energy and time

Well suited to solving complicated three-dimensional, time-dependent problems

The use of the Monte Carlo method as a radiation transport research tool was started at Los Alamos National Laboratory during 1940s

We will be using two MC based codes: MCNP and TRIM in our tool to simplify the complexity

Terrestrial Cosmic Ray Spectrum

SESEE:Flow Diagram

Neutron Si interaction

Ion transport

Qcollected

Circuit: Netlists &

Layouts

Qcritical

SER

Terrestrial Cosmic Ray Spectrum

SESEE:Flow Diagram

Neutron Si interaction

Ion transport

Qcollected

Circuit: Netlists &

Layouts

Qcritical

SER

NEUTRON TRANSPORT

A general-purpose, continuous-energy, generalized-geometry, time-dependent, coupled Monte Carlo N–Particle (MCNP) transport code Used for neutron, photon, electron, or coupled neu

tron/photon/electron transport, including the capability to calculate eigenvalues for critical systems

NEUTRON TRANSPORT

MCNP Uses a continuous energy scheme, rather than

energy groups Neutron energy range: 10-11 MeV to 150 MeV photon and electron energy range from 1 keV to 1 GeV

Has generalized 3-D geometry capabilities with elaborate plotter capabilities

Has elaborate tally capabilities “Ptrac” gives elaborate details of the neutron transpor

t location, collision, absorption for each neutron

MCNP

We get the different collisions and there positions in the silicon Can be aborption Or Inelastic scattering creating charged particles like

Carbon, Al or Mg etc Currently each resultant reaction is assumed to be of the equ

al probability, we plan to augment the tool through the use of the more detailed codes like MECC

MCNP

The reactions modeled are

The sample cell contained both P-type (containing Boron) and N-type regions (containing Phosphorus)

MCNP Model

Side View of the model

Top View of the model

Side View of the model

B: p well

P: n well

Neutron Transport: Results

The MCNP input is the terrestrial cosmic ray PDF

The MCNP output is from the ‘ptrac’ card The Ptrac result is parsed and converted into a

form that TRIM can accept as input For 300 million neutrons injected,

There are 1100 inelastic scattering event A 10 absorption events

Terrestrial Cosmic Ray Spectrum

SESEE:Flow Diagram

Neutron Si interaction

Ion transport

Qcollected

Circuit: Netlists &

Layouts

Qcritical

SER

Ion Transport

Neutron collision or absorption may be followed by charged particle creation

Charged particles create random electron hole pairs and get collected at junction

The process can be modeled by Monte Carlo We use TRIM to simulate the transport of

charged particles TRIM calculates the ion tracks and the

resultant ionization

Ion Transport

On plotting the ionization and the particle penetration depth (Bragg’s Curve)

Resultant ionization is assumed to give a point charge

The results are stored as a table

Typical Results using TRIM

The typical reaction byproducts 4He, 12C, p and n create the ionization of the range of 2.5E4e-h/um, 4E5 e-h/um, 3.7E3 e-h/um, 0

These correspond to a Charge of 3.5e-13C, 1.8e-13C, 2.321e-13C

Terrestrial Cosmic Ray Spectrum

SESEE:Flow Diagram

Neutron Si interaction

Ion transport

Qcollected

Circuit: Netlists &

Layouts

Qcritical

SER

Terrestrial Cosmic Ray Spectrum

SESEE:Flow Diagram

Neutron Si interaction

Ion transport

Qcollected

Circuit: Netlists &

Layouts

Qcritical

SER

Circuits

At circuit level Qcritical is the most common metric to evaluate the robustness

Evaluated using circuit simulators such as Hspice.

Transient Pulse modeled as current pulse with a sharp rise and slow decay

The measured Qcritical is stored in the file along with the node name as a table

The input to the tool are the GDSII files with the different node names and the tabulated Qcritical values

Circuits

0

Circuits

GDSII GDSII is a binary file format which is classified as a

"data interchange format", used for transferring mask-design data between the IC designer and the fabrication facility

GDSII gives the spatial information unlike spice netlists which account for only electrical properties

Effective in evaluating the schemes such as interleaving and their effect on soft errors. Hence, the spice netlists accounts for the process, device

and circuit while GDSII files of the circuits account for topology

Circuits

Charge collection and Funneling depth are given by the TRIM simulation which give the Qcollected

The Qcollected is weighed based on the depth D and the distance from the circuit node.

The GDSII files are parsed using the xrlcad2.0 library package

An failure is assumed of the ratio of the weighed Qcollected/ Qcritical lesser than 1

Circuits

Charge Sharing Past studies have shown the charge sharing to

be additive or linear for related nodes For non related nodes and inverse square rule

is used to explain the charge sharing. Such an approximation is used to reduce the

computation time as the alternate of simulating the charge transport in silicon is very time tedious and requires extensive device level simulations.

Case Study

A Custom designed SRAM is chosen The Qcritcal for the two nodes was calculated

by simulating in Hspice, and found as 15 and 330fC

So for a typical inelastic scattering, the Soft Error will observed from only those particles that generate a secondary 4He particles.

Further work

Effect of charge depth has to be modeled Make use of Monte Carlo based nuclear

internal cascade codes for precise modeling of inelastic scatterings