Presentation is loading. Please wait.

Presentation is loading. Please wait.

M. Aguirre1, J. N. Tombs1, F. Muñoz1, V. Baena1, A. Torralba1, A

Similar presentations


Presentation on theme: "M. Aguirre1, J. N. Tombs1, F. Muñoz1, V. Baena1, A. Torralba1, A"— Presentation transcript:

1 FT-UNSHADES: a new system for SEU injection, analysis and diagnostics over post synthesis netlists
M. Aguirre1, J.N. Tombs1, F. Muñoz1, V. Baena1, A. Torralba1, A. Fernández-León2,, F. Tortosa-López2 1Escuela Superior de Ingenieros Universidad de Sevilla Camino de los Descubrimientos s/n Sevilla (SPAIN) 2Data Systems Division. ESTEC/TOS-ED European Space Agency. Noordwijk (THE NETHERLANDS)

2 MAPLD 2005, Washington (USA)
Summary Brief Description of FT-UNSHADES project What is not FT-UNSHADES? FT-UNSHADES secret: Partial Reconfiguration The system Design flow Tests procedure Analysis procedure Benchmarking Conclusions The Future Q&A MAPLD 2005, Washington (USA)

3 Brief Description of FT-UNSHADES project
Radiation environment emulator Use a huge FPGA from Xilinx Initial objective: Provide a test and analysis framework to check the design protections (TMRs and voters, EDACs, …) of an IP or ASIC, BEFORE place and route and fabrication Current solutions: VHDL simulators Instrumenting FFs 1 SEU D Q MAPLD 2005, Washington (USA)

4 What FT-UNSHADES isn’t?
It’s not a platform for radiation testing of FPGAs It’s not designed for being inserted into a radiation environment MAPLD 2005, Washington (USA)

5 FT-UNSHADES secret: Xlinx partial reconfiguration
Takes advantage of partial reconfiguration techniques for fault injection and capture and readback for design observation Changes in FF state are made from the configuration circuit. Netlist does not need to be modified to pass FT-UNSHADES fault-injection tests. Direct manipulation of bitstream slices speeds-up the bit-flip induction. Large circuits with huge stimulus vector database are tested at hardware speed CIRCUIT IS TREATED AS A BLACK BOX MAPLD 2005, Washington (USA)

6 System elements: the board
Vector Memories 2Mx102 Multi Board Link SYSTEM FPGA Virtex II Control FPGA Spartan II-50 PC-Host USB Link MAPLD 2005, Washington (USA)

7 MAPLD 2005, Washington (USA)
System elements: the software TNT INPUT VECTOR DATABASE BITSTREAM BIT ALLOCATION User Commands (Scripts) WAVEFORM Session.log Excel Sheet Console MAPLD 2005, Washington (USA)

8 MAPLD 2005, Washington (USA)
Test Model Two identical instances of the Module Under Test are synthesised with the Test Shell: -SEU_MUT receives the bit flips -GOLD_MUT allows comparisons Both MUTs must be treated as Macros, to avoid their synthesis collapsing. Test Shell: -R/W the memories -Controls the clock -Reports events -Controls the display MAPLD 2005, Washington (USA)

9 MAPLD 2005, Washington (USA)
Design flow MAPLD 2005, Washington (USA)

10 Test Procedure: When, Where and How
Configure the System FPGA and download the input vector database Specify the test level (How): Damage Output Latent Specify time (When) Specify location (Where) Specify method and conditions (How) RUNTEST <- Fault Dictionary is generated MAPLD 2005, Washington (USA)

11 MAPLD 2005, Washington (USA)
A single test process: SEU(T,L) A single test run has the following schedule: Program the time counter with T Give Module Reset Launch the system to stop at T Get frames related to L Induce SEU at L Resume system Watch events to check SEU effects. MAPLD 2005, Washington (USA)

12 MAPLD 2005, Washington (USA)
Analysis Procedure: Step: Run a single clock cycle and check internal evolutions AnRun Command: Run the emulation to the insertion point. Then analyse. Diff command: Compare the states of Gold/SEU instances to check internal propagation MAPLD 2005, Washington (USA)

13 MAPLD 2005, Washington (USA)
Benchmarking MAPLD 2005, Washington (USA)

14 MAPLD 2005, Washington (USA)
Conclusions Circuit Overhead: system gates Fault injection time: 0,8ms (XC2V6000) and 0,9ms (XC2V8000) A tool for verification of your design protections Automatic search of weak/unprotected points Understand/reproduce rad test results Analysis Non-intrusive controlled (Time,Location) SEU injection If you want to reduce the protection level … or optimize critical areas MAPLD 2005, Washington (USA)

15 MAPLD 2005, Washington (USA)
Future of FT-UNSHADES Large circuit benchmarking (Leon2) Output files for automatic TMR insertion tools, in order to produce selective protection Test with multi SEUs Improve system performance Extend the injections to memory blocks MAPLD 2005, Washington (USA)

16 Thank you for your attention
Q&A MAPLD 2005, Washington (USA)


Download ppt "M. Aguirre1, J. N. Tombs1, F. Muñoz1, V. Baena1, A. Torralba1, A"

Similar presentations


Ads by Google