Center for RC eliable omputing What Defect Does L Testing Detect? Stanford University Nov. 1, 1999 Outline Introduction Physics of Tunneling Proposal of Tunneling Defect Theoretical Calculation Experiment Evidence Future Work Summary 1 Introduction L effectiveness table [Chang 96] Failure Mode IDDQ L Transmission gate opens NO YES Threshold voltage shifts NO YES Diminished-drive gates NO YES Gate oxide shorts YES YES Metal shorts YES YES Defective interconnect buffers YES YES High resistance interconnects NO NO Tunneling open NO NO Introduction 9 L-only CUTs in Murphy 5 CUT can not be explained by the table Goal of this talk 3 4 Physics of Tunneling Put Them ll Together 5 6 Page 1 Nov.1, 1999
Proposal of Tunneling Defect L-only CUT behavior Previous study Possible explanations Proposal of a tunneling defect formation of a tunneling defect qualitative description of the circuit behavior at nominal voltage at very low voltage 6 5 4 3 1 L-only CUT Behavior Long time IDDQ leakage phenomena IDDQ in u 7.1.7.13.18.4 time (second) 8 Maly Previous Study Possible Explanations location C R Logic 1fF 1T 1pF 1G I/O 1pF 1G 9 1 Formation of a Tunneling Defect Qualitative Description I DDQ (t) very long leakage Metal X SiO SiO Poly very thin in out 11 1 Page Nov.1, 1999
@ Nominal oltage @ ery Low oltage fast ckt response log I very slow I DDQ leakage log I very slow response very slow I DDQ leakage 5 13 1.7 14 Theoretical Calculation electrical model analytical equations simulation results nominal voltage L IDDQ leakage J tunnel Electrical model d, = defect oxide thickness, area E ox = ox /d ox C d metal C gate poly 15 16 nalytical equations Tunneling effect (FN+Dir) [Schuegraf 9] tunnel Eox J ( t) = e correction _ fector = (1 ( / Eox ) b b ox ) e B( b ox ) b / E ox Simulation: 5,C gate = 4fF, d=1 =1 m 6 5 4 m p_tunnel p_couple J poly( t) = C Coupling Effect tunnel gate ( t) dt voltage 3 = ox d Cd poly = metal Cd + C Cd gate 17 1..4.6.8 1 1. 1.4 1.6 1.8 time (ns) 18 Page 3 Nov.1, 1999
I DDQ Leakage Simulation: 1.7 nalytical equation voltage 6 5 4 3 1 m p_tunnel p_couple..4.6.8 1 1. 1.4 1.6 1.8 time (ns) 19 T leakage = C J gate L_tunnel ssumptions (@ nominal voltage) = (th -t) =.5 -.7=1.8 C gate = 4fF T leakage = 1ms ~ 1S = 1 m Calculation result J L_tunnel = 7 ~ 7. /cm Experimental Evidence Boolean Test - Expected Behavior Boolean Test Evidence expected behavior experimental results IDDQ Test Evidence expected behavior experimental results Failure nalysis Evidence If not tunneling [Chang 96] Delay Ratio = T_bad / T_good either fail or less than 1 x slower Defect Max Delay Retio @L Weak driven gate 41.7 Tran. gate open Fail t shift.98 Diminished drive gate 3.3 1 Boolean Test - Expected Behavior Boolean Test - Expected Behavior If Tunneling extremely long delay @ L eventually work minimum functional voltage same as good Expected shmoo plot N Good not tunneling tunneling L min 3 ns s ms delay 4 Page 4 Nov.1, 1999
Boolean Test - Experiment Results Table of delay time @ different voltage I DDQ Test - Expected Behavior Table of expected IDDQ behavior cut id 5..5. 1.7 Good 3nS 6nS 1ns 47nS 13.69.6sq 38nS 114nS N 6nS 31.9.6sq 3nS 86nS N Fail 35.8.m1 36nS N Fail Fail 5.15.6sq 36nS N Fail Fail 1.3.m1 34nS 3ns 5 S 3mS 1.91.6sq 37nS 148nS 4nS 1 S 18.133.m1 35nS ns 3 S ms 1.34.6sq 37nS 9nS 6 S 3 S 35.83.m1 35nS 7nS 9 S 4 S defect high IDDQ () IDDQ leakge GOS Y N - - t shift N N - - WD gate Y N - - Tr. gate open Y N - - RC Y Y S No D/S junction leakage Y Y ms ~ S decrease decrease Tunneling Y Y ms ~ S No 5 6 IDDQ in u I DDQ Test - Experiment Results I DDQ leakage @ different voltage (from 1.91.6sq) 8 7 6 5 4 3 1.5.1.15..5.3 Time (second) 5 6 3.5.5 7 I DDQ Test - Experiment Results Table of experimental I DDQ behavior cut id max decrease leakage IDDQ decrease 13.69.6sq N - - 31.9.6sq N - - 35.8.m1 18 N - - 5.15.6sq 38 N - - 1.3.m1 88 Y 14~56mS N 1.91.6sq 5 Y 56~7mS N 18.133.m1 14 Y 8~69mS N 1.34.6sq 35 Y 4~57mS N 35.83.m1 1 Y 4~55mS N 8 Failure nalysis Sematech data [Nigh 98] dd=5 dd=3.3 dd=1.8 m d t ns m d t ns m d t ns 1 1 1.3 11.6.7 1.5 9 3 Page 5 Nov.1, 1999
Q: Why sequence dependence @ L? ll tunneling defect CUTs are seq. dept. @1.7 tunneling effect has polarity rise time different than fall time Q3: Is it process error or defect? Defect 5 CUTs from 3 lots, different wafer Q4: What is the best test condition? L transition pattern maybe low temperature 31 Q5: Will it cause reliability problem? Yes low quality oxide, break down easily No Sematch data [Nigh 98] IDDQ did not change much after burn-in after break down, becomes high impedance 3 Q6: Will it cause more trouble in the future technology? No Cg smaller, coupling effect dominates dd decreases, FN tunneling decrease Yes metal intensive direct tunneling is field depdent 33 Summary 5/9 L CUT can be explained by FN tunneling CUT ID high IDDQ slow IDDQ leakage @L possible defect 13.69.6sq N N N t shift? 31.9.6sq N N N t shift? 35.8.m1 Y N N shorts? 5.15.6sq Y N N shorts? 1.3.m1 Y Y Y Tunneling 1.91.6sq Y Y Y Tunneling 18.133.m1 Y Y Y Tunneling 1.34.6sq Y Y Y Tunneling 35.83.m1 Y Y Y Tunneling 34 Reference ppendix 35 Page 6 Nov.1, 1999
Direct Tunneling Oxide thinner than 4~5nm Fowler-Nordheim Tunneling Strongly depends on electric field 37 38 Trap ssisted Tunneling Depends on oxide quality I DDQ Test - Experiment Result Definition of I DDQ I(t) = I leak exp (-t/ ) + I final I leak I fianl 39 time 4 Page 7 Nov.1, 1999