• Aucun résultat trouvé

Accurate Modeling of Fault Impact in Arithmetic Circuits

N/A
N/A
Protected

Academic year: 2021

Partager "Accurate Modeling of Fault Impact in Arithmetic Circuits"

Copied!
2
0
0

Texte intégral

(1)

HAL Id: hal-01404772

https://hal.inria.fr/hal-01404772

Submitted on 29 Nov 2016

HAL is a multi-disciplinary open access

archive for the deposit and dissemination of sci-entific research documents, whether they are pub-lished or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers.

L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.

Accurate Modeling of Fault Impact in Arithmetic Circuits

Pierre Guilloux, Arnaud Tisserand

To cite this version:

Pierre Guilloux, Arnaud Tisserand. Accurate Modeling of Fault Impact in Arithmetic Circuits. DASIP: Conference on Design and Architectures for Signal and Image Processing (Demo Night), Oct 2016, Rennes, France. 2016. �hal-01404772�

(2)

Accurate Modeling of Fault Impact in Arithmetic Circuits

Pierre Guilloux and Arnaud Tisserand.

IRISA – CNRS – Univ. Rennes 1

Reliasic

1. Objective: Design of Robust Arithmetic Circuits

Study links between fault detection/tolerant solutions and

I types of operations ±, ×, ×cst, ÷, √, sin, cos, log, . . .

I computation algorithms and representations of numbers

I optimizations at architecture and circuit levels

I mathematical error due to faults and rounding modes

Applications: signal/image processing, embedded computing,

protections for cryptrographic devices, . . .

2. Intensive Logical Fault Simulations: Emulation on FPGA

cells from ASIC library injection controller

x y =      x ¯ x 0 1 sel logical fault injector serv er side FPGA side HDL block ASIC tools gate level netlist instrumentation (add injectors) instrumented netlist FPGA tools bitstream FPGA EMULATION 3. Hardware Platform server

FPGA 1 FPGA 2 FPGA N

LAN (switch) GbE links power supply power cables User network 1 server 8 cores 4 GHz 1 power supply 1 GbE switch 4 FPGA Zedbord 7020 1 FPGA PicoZed 7035 1 FPGA Zybo 7010 Total cost 3800 eHT

4. Emulation Engine(s) in FPGA

reference block BUE PI distribution module on-line analysis module interface injection CTRL main CTRL x y ˜ y ARM core ARM core AXI bus emulation engine 1 emulation engine 2 PS PL

reference operator computes y = f (x )

faulted operator computes y = ˜f(x )e

mathematical error analysis ε = |y − ˜y |

5. Tools Flow

FPGA 1 FPGA 2 FPGA N

cell library ASIC tools FPGA tools bitstream definition step block emulation directives serv er side FPGA side Platform preparation

step emulationstep

results analysis

Possible locations for logical fault injection:

I manual on selected internal nets (for debug)

I random on internal nets

I random over circuit area (based on cells area)

6. Example: Multiplier with Residue Code Detection

16×16-bit integer multiplier, moduli m ∈ {3, 7, 15}, with

(a × b) mod m = (a mod m) × (b mod m)  mod m

step duration area delay

Impl. ASIC orig. mult. 25 s 1164 gates –

Impl. FPGA orig. mult. 43 s 444 LUT 9.4 ns

Impl. FPGA instrum. mult. 48 s 453 LUT 10.1 ns

Impl. FPGA emul. engine 389 s 1714 LUT + 1 BRAM 11.0 ns

boot petalinux Zedboard 15 s – –

emulation 1 fault & 1 EE 94.5 s – –

Comparaison to a CABA simulation in software (Vivado 2014.4) :

×6880 speedup for 1 emulation engine

Error distribution for exhaustive tests (232 inputs) and various

sin-gle faults (1 color = 1 fault) :

0 20 40 60 80 100 0 5 10 15 20 25 30 35 40 frequency [%]

error [bin index in 0..255]

Residue code implementation results:

modulo area delay mesured avg. detection rate

m [LUT] [ns] [%]

3 525 10.4 66.7

7 526 10.1 85.6

15 553 13.0 93.3

7. Future Works

100 EEs @ 100 MHz / second / hour / day

nb. faults or PIs 100 · 106 × 100 ≈ 233 ≈ 245 ≈ 249

References:

Arithmetic operators library from R. Zimmermann (http://www.iis.ee.ethz.ch/ zimmi/arith_lib)

P. Guilloux & A. Tisserand, Compas 2016, French Conf. on Parallelism, Architecture and System

Références

Documents relatifs

We can see that when the photovoltaic production increases and is superior to the consumption, the agent will switch the position to 1 and take this source of energy

Ce principe utilisé pour les mémoires, l'est aussi dans les CPLD (Complex Programmable Logic Device), mais surtout dans les FPGA (Field Programmable Gate Array) sous

A realistic hardware emulation of the brain, the Synthetic Neuro-anatomy is capable of learning and matching learned waveform patterns in time (temporal) and

The Compare-Add-Increment (CAI) architecture adopts some features from the carry-increment adder, a widely adopted structural simplification of the carry-select scheme. In

The paper “A 115,000-year-old expedient bone technology at Lingjing, Henan, China” by Doyon and colleagues presents some quantitative and qualitative criteria to identify

After describing the hard- ware platform on which the microkernel has been ported, We will focus on the proposed microkernel and the custom specific system task dealing with

On this platform, high speed IO device is connected to an FPGA-based IO controller through optical fiber and the IO controller talks to the FPGA-based synthesizable processor core

The reverse engineering of the input/output block has three goals: (i) map a number of the configuration options for each pin to their respective indices and values in the bitstream