Reliability of TSV interconnects in 3D-IC

Transcription

1 Reliability of interconnects in 3D-IC Electromigration voiding analyzed through 3D-FIB-SEM T. Frank, C. Chappaz, F. Lorut - STMicroelectronics, Crolles, France P. Leduc, L. Arnaud, S. Moreau, A. Thuaire - CEA-LETI, Grenoble, France L. Anghel - TIMA Grenoble INP, Grenoble, France "Challenges for three-dimensional (3D) ICs and Systems", Workshop, November 28-29, Toulouse, France

2 Context & Purpose Reliability of is a major concern 15µm Metal Top (Cu) Characterize electromigration of 3D interconnects Metal Top 2.3µm (Si) (Cu) Metal Bottom (Cu) e Metal Bottom Propose a model of electromigration behavior Supported by Failure Analysis 2

3 Outline Introduction Context & Purpose Electromigration in usual Cu BEOL interconnects Experimental technology Test structure Resistance increase Failure analyses Model of electrical resistance increase Model Parameter extraction Discussion Electromigration robustness bottom section Conclusion 3

4 Electromigration in usual interconnects Electromigration Cu migration driven by j[ma/cm²] & T[ C] a Via Void e- Voids and Hillock formation in usual Cu BEOL Interconnects Via-ended line: ~2MA/cm² & ~300 C b Line 65nm node Void Monitoring of R[Ω] of a via ended line e- c Resistance e- c Void e- b t a time L. Doyen, JAP

5 Through Silicon Via technology Metal Top Stack 8 inches WoW Direct Bonding Cu, Via last Height = 15µm Diameter = 2.3µm SiO2 & TiN integration 15µm 2.3µm Si Metal Bottom SiN SiO2 Interconnects Cu Cu TiN Metal bottom Cu Damascene SiO2, SiN, TiN integration Line/ interfaces: Cu/TiN/Cu 5

6 2. 3µ m Electromigration test structure V- I+ 4µ m I- V+ 6µ m e500µm 6µm 4 x Metal Top e 4 x Si Metal Bottom Stress Conditions: 15/25mA & 270/300 C 6

7 Monitored resistance trace Electrical resistance increase: Latency period at start (1) Logarithmic increase (2) No sudden step up Different from R(t) of electromigration in usual Cu interconnects Stress conditions: 15-25MA/cm² & C 7

8 Failure Analyses 2D FIB SEM Aggressive form factor (15µm / 2µm) (Si) Large FIB windows Risk of missing structure & failure site Density mismatch of Cu, Si, SiO2 relief transfer Time : ~2 hours No automation No void volume data Benefit Large window top & bottom interfaces Accurate localization of electromigration void Bottom interface Top interface 8

9 Failure Analyses 3D FIB SEM Aggressive form factor (15µm / 2µm) Sample preparation Reduced window size (need of pre-2d FIBSEM) Time : ~6 hours Benefit No Risk of missing structure & failure site Automated Reconstructed pictures Accurate localization of electromigration void Volume computations Sample preparation (reduce window size) (Si) FIB direction Metal level void bottom section Reconstructed 3D view Reconstructed Plan view 9

10 Failure Analyses Voiding vs. time µ 2.3 ΔR void m Bottom line e- +8,7Ω <+0.1Ω X-section Plan views 76h m 4µ 4,4h ΔR e- void t e- t ebottom line void e bottom section evoid bottom section 10

11 Model of resistance increase FIB SEM failure analyses Model of: Electromigration physics R[ ] f (time ) - Void grows right under the - Radial void growth - Void spans over whole line thickness 1µm bottom section Plan view bottom section Plan view tcu rvoid Cross section view 11

12 Model of resistance increase Model of resistance increase: t R(t ) R0 A ln( ), t t 0 t0 R(t ) R0 0, t t 0 Cu line thickness Proportional constant A Barrier thickness radius 2 Electromigration Cu flow Model is fitted for each sample, and (A, t0) are extracted t0 Ln( t [h] ) B A 4 tb t Cu r t0 F xp trace model ΔR[Ω] Barrier resistivity 200 Model fits on 89 out of 128 tested samples 12

13 Parameter extraction t R(t ) R0 A ln( ) t0 xp trace model ΔR[Ω] FC (10 %) A t0 2 t Cu r t0 F Electromigration TTF Ln( t [h] ) Current expon. MTF A j n Current density 200 Compare Activation energy EA t 0 j exp( ) k BT EA exp k BT Temperature Black s law: MTF n, EA MTF: TTF: Cu flow Mean Time To Failure Time To Failure n F (current exponent), EA(activation energy). 0 13

14 Parameter extraction Compare (n;ea)t0 vs. (n;ea)ttf But t0 and TTF extractions have to be independent Time To Failure (TTF): Failure Criterion:10% TTF 10% t0 M od el fi t ΔR [Ω] Point of resistance trace. ln ( t [h] ) t0 : Evolution of resistance increase. 14

15 Parameter extraction ctivation Energy (EA): Model s t0 extraction C 25mA Black s Law (TTF) 90 EA 250 C 25mA 100 t0 [a.u.] EA,t0 = 0.9 ±0.2 ev. 300 C 50 25mA 10 EA 250 C 25mA Time to Failure [a.u.] EA,TTF = 0.9 ±0.1 ev. 15

16 Parameter extraction urrent Exponent (n): Model s t0 extraction mA 300 C Black s Law (TTF) 90 -n 50 15mA 300 C 100 t0 [a.u.] nt0 = 1.8 ± mA 300 C -n 15mA 300 C Time to Failure [a.u.] nttf = 2.0 ±0.1 n & EA extracted values are relevant regarding values calculated from Black s equation using Time To Failure 16

17 Parameter extraction Summary: Extracted values of current exponent and activation energy match independent experimental values Model of resistance increase matches experimental data Extract guidelines to increase robustness of ended interconnect 17

18 Electromigration robustness bottom section The instant the resistance starts to increase depends S 2 t Cu r k BT n EA t0 i exp( ) * N i Deff0 Z Cu k BT To increase robustness : Increase section [Process / Technology] No benefits from pad or line width increase [Design] Possibility to add redundant [Design] S WPad WLine 18

19 Conclusion & Question Propose a model of electromigration in a ended line Experimental observations Electrical resistance increases under T[ C] and i[ma] stress FIB SEM analyses reveal void grows in bottom line under A model is proposed for the resistance increase Radial growth of void under Model enables extraction of aging parameters (EA and n) Propose solutions to increase robustness Workshop Question Failure analyses are essential for 3D-IC reliability studies FIB SEM analyses are limited by the dimensions Possible to use non destructive analyses? 19

20 Thank You for your attention

21 Annex 21

22 Model of resistance trace Purpose: Give an expression of: R[ ] f ( t ) Based on void growth kinetic Assumption 1: No Matter Back-Flow Assumption 2: Radial void growth Assumption 3: bottom section approximated to a circle

23 Hypotheses of the model Assumption 1: No Matter Back-Flow F FEM FBF (t ) Total matter Flow: EM L BF L 1 FBF (t ) FEM jstress 1 EM FEM: Electromigration matter flow FBF: Backflow induced by gradient of matter concentration VM: Depleted matter volume F FEM t VM (t ) F dt VM (t ) F t Constant Flow during test : 0

24 Hypotheses of the model ΔR Assumption 3: bottom section is circular Bottom Line e- Void larger than bottom section: Electrons have to flow through the barrier void R(rVoid ) 0 e- t

25 Hypotheses of the model ΔR Assumption 3: bottom section is circular Bottom Line e- tb void r rvoid t Void larger than bottom section: Electrons have to flow through the barrier Approximation of the circular bottom section enables direct expression of R[Ω] increase B rvoid R(rVoid ) ln( ) 2 tb r e- ΔR: Resistance increase r: radius rvoid: Void radius radius tb: Barrier thickness