Pro Asic3 - Radiation test at CHARM Christophe Godichal – BE/BI/QP 1
Test Setup 2
40 GBTx Elinks 80 lines in the FPGA 2 use for Reset, Enable flag 78 use for the test logic 3
FPGA Configuration Test Logic used 52% of the FPGA 4
Test Logic 16 Shift-Register with different combinaison of logic gate between register Without logic gate between registers NOT gates AND gates OR gates 12 Shift-Register TMR with the same combinaison of the Shift-register without TMR 5
Test Logic: Shift-Register Objective: SEU detection Register arrangment 2 SR with 350 reg. placed « randomly » by the compiler 3 SR with 350 reg. placed manually at the extreme of the FPGA 32 bit shift register (manual placement) Long connection in the fabric 6
Shift-Register with Logic Gates Shift-Registers with 8 NOT gates between each register Objective SEU detection SET 2 SR with 350 reg. placed « randomly » by the compiler 3 SR with 350 reg. placed manually at the extreme of the FPGA Shift-Registers with 8 AND gates between each register Sensibility of SEU and SET 3 SR with 350 reg. placed manually at the extreme of the FPGA Shift-Registers with 8 Or gate between each register Sensibility of SEU and SET 3 SR with 350 reg. placed manually at the extreme of the FPGA 7
Shift-Register With « TMR » Shift-Registers with « TMR » SEU Immune, SET in the voter 3 Simple SR, 3 with 8 NOT gate, 3 with 8 AND gate, 3 with 8 OR gate (all SR with 350 reg TMR) Test is used to compare sensibility of Shift-Register with TMR and without TMR 8
Results - TID results Type of SR Manual Placement in the FPGA Placed by the CompilerShift-Register TMR Gy SR_ SR_ SR_ SR_NOT_ SR_NOT_ SR_NOT_ SR_AND_ SR_AND_ SR_AND_ SR_OR_ SR_OR_1 459 SR_OR_2 485 Manual placement in the FPGA has failed from 485Gy to 538 Gy Placed by the compiler has failed from 467Gy to 492Gy Shift-Register TMR has failed from 461Gy to 511Gy FPGA Stopped working at 752Gy 9
Results #errors and Cross-Section Type Total Nb of Register Total Nb of Errors Cross- Section Cross Section uncertainty Simple E E-15 Not Gate E E-15 And Gate E E-15 Or Gate E E-15 Without TMR 10
Results #errors and Cross-Section With TMR Type Total Nb of Register Mean (λ) std dev (σ) Cross- Section Cross Section uncertainty Simple E E-15 Not Gate E E-15 And Gate E E-15 Or Gate E E-15 11
The I/O of the Shift-Register with TMR are not TMR too We have 2 register in the input and 1 register in the output port What is the probability to have one error on these i/o port ? Results Confidence on TMR results Type#I/O reg Cross- Section Error I/O reg probability Simple39.14E Not Gate38.76E And Gate38.33E Or Gate38.90E
Results Difference between SR not TMR and SR TMR Type No TMRTMR Improvement σσ Simple9.14E E Not Gate8.76E E And Gate8.33E E Or Gate8.90E E
Conclusion All shift register stopped around 500Gy The ProAsic dies around 750Gy We have few errors on TMR-ed chain, so the statistics or not really good We can suppose error in the I/O cell in the Shift-Register TMR Future testing can be programmed to test more in details the propagation delays 14