Keeping Your CMP Slurry From Being A Pain in the As-Probed Die Yield. Robert L. Rhoades (Entrepix) Brian Orzechowski and Jeff Wilmer (DivInd, LLC)

Transcription

1 Keeping Your CMP Slurry From Being A Pain in the As-Probed Die Yield Robert L. Rhoades (Entrepix) Brian Orzechowski and Jeff Wilmer (DivInd, LLC) Presentation for the Levitronix Conference February 1, 29

2 Outline Background Generalized Diagnostics Examples Summary 2

3 Background Components of successful CMP Polisher+pad+slurry+conditioning+wafer+film+luck Process has been in manufacturing for >15 yrs Excursions and deviations still occur Control limits on most wafer metrics keep shrinking Downtime is costly scrap is even more costly! Slurry is a key factor for all major CMP processes Removal rate, selectivity, roughness, dishing, erosion, defect density, etc. can all be affected by slurry Storage and distribution are critical 3

4 Why is this important? Higher large particle counts (LPC) = higher defects Particle Counts vs. Mirra 4 Particle Count Particle Count Global POU Micro Scratch Microscratch Counts Slurry properties have a DIRECT connection to polished wafer defect metrics Feb 27-Feb 1-Mar 3-Mar 5-Mar 7-Mar 9-Mar 11-Mar 13-Mar 15-Mar Date Downtime and scrap both have a high cost. 4

5 Fishbone Diagram Things that make you go HMMM FAILURE Human Machine Materials Methods A helpful brainstorming tool Results easily transfer to FMEA (if desired) 5

6 General Sequence Failure A familiar sequence to any fab process engineer CMP Process Off Line Diagnostics & Troubleshooting Does process pass qual? Yes Release to Mfg The key is to find root cause and get back on line as quickly as possible! No 6

7 Diagnostic Sequence Single Tool Fix and Verify End of pad life Conditioner life Filter (if used) Peristaltic tubing Calibration drift Valve (post-loop) Pump (if present) 7

8 Example #1 Observations Toolset running stable One idle polisher was brought back on line and failed defect quals on successive tries LPC tail shows delta between slurry loop and sample at platen Number Of Particles >.56 Microns Mirra 2 Points F/G-5 Minute Test Point G Point H Solution Sample Time (minutes) Perform PM on tool Returned to baseline so further action not required

9 Diagnostic Sequence Single Tool Multi-Tool Commonality Fix and Verify End of pad life Conditioner life Filter (if used) Peristaltic tubing Calibration drift Valve (post-loop) Pump (if present) Sudden onset? Slurry lot change Loop filter change Test wafer lots Operating setpoints Pumps Valves 9

10 Example #2 Oxide CMP Rate qual failure Simultaneous shift in uniformity Oxide CMP Removal Rate & Uniformity Mirra, 2k thermal oxide Removal Rate (Ang/min) Uniformity (% 1-sigma) Observations Series of similar qual fails on multiple tools No shift in defects Rate Rate-UCL Rate-LCL WIWNU% WIWNU-UCL

11 Solution #2 Diagnostics Label possibly related events on chart Clear timing with new slurry lot (new tote) Short Term Fix Purge / flush / refill Recharge with a different slurry lot Long Term Improvements Removal Rate (Ang/min) Pad chng Oxide CMP Removal Rate & Uniformity Mirra, 2k thermal oxide Pad chng Rate Rate-UCL Rate-LCL WIWNU% WIWNU-UCL Improved control at slurry manufacturer In-line monitoring for ph and S.G. (% solids) Pad chng Slurry tote chng Uniformity (% 1-sigma) 11

12 Example #3 Observations Drum, Silica, Mondisperse Particle monitor installed for passive data collection Small random spikes in 2um and 5um bins correlate with wafer level defect qual data No commonality to tool, pad changes, etc. Coincided with a fraction of drum changes Counts (See Legend) /29/99 : 7/4/99 : Incident #1 June 8 July July 21: 21: High High PSD PSD counts counts from from drum drum in in agreement agreement with with Defectivity Defectivity 7/9/99 : Date of Sample Incident #2 July 21 7/14/99 : Incident #3 August 2 7/19/99 : 7/24/99 : Particles Particles Particles Particles Particles = Post Bin 1 ( µm) = Post Bin 2 (.26-.5µm) = Post Bin 3 (.5-1.6µm) UCL POSTBIN1: 1795 POSTBIN2: 14 POSTBIN3: 3 12

13 Solution #3 Diagnostics Loose commonality to certain lots of slurry Filtration tests promising Short Term Fix Purge / flush / refill Transfer filter Continue monitoring Long Term Improvements LPC and defect qual data correlation confirmed Early flag for engineering on any OOC data point at transfer Particle Counts (#/ml) Bin micron Bin micron Bin micron 13

14 Diagnostic Sequence Single Tool Multi-Tool Commonality Trend Analysis Fix and Verify End of pad life Conditioner life Filter (if used) Peristaltic tubing Calibration drift Valve (post-loop) Pump (if present) Sudden onset? Slurry lot change Loop filter change Test wafer lots Operating setpoints Pumps Valves SLURRY PARMS Slurry ph Density (or S.G.) Concentration [X] WAFER DATA Rate / Uniformity Defects (qual) Defects (on product) SERVICE OPERATIONS System settings Pump rebuild life Batch transfers Drum switchovers 14

15 Example #4 Tungsten CMP Random qual failures Some recovery after pad changes Tungsten CMP Removal Rate & Uniformity Mirra, 2 mm blanket film W wafers Removal Rate (Ang/min) Uniformity (% 1-sigma) Observations Observed on multiple tools No shift in defects Rate Rate-UCL Rate-LCL WIWNU% WIWNU-UCL

16 Solution #4 Diagnostics Plot with trend line Assay slurry [H2O2] Fresh mix In loop Short Term Fix Purge / flush / refill Manual [H2O2] monitor Removal Rate (Ang/min) Tungsten CMP Removal Rate & Uniformity Mirra, 2 mm blanket film W wafers Uniformity (% 1-sigma) Long Term Improvements Avoid excess day tank volume (keep turnovers reasonable) In-line monitoring for [H2O2] and auto-dose replenishment Rate Rate-UCL Rate-LCL WIWNU% WIWNU-UCL Power (Rate) 16

17 [H2O2] Decay Bench Test Single batch of tungsten slurry Target mix 3% H2O2 Circulated in clean loop with data point taken every 1 min. H2O2 Concentration (wt%) Result 1/12/9 3:36 AM 1/12/9 4:48 AM 1/12/9 6: AM 1/12/9 7:12 AM 1/12/9 8:24 AM 1/12/9 9:36 AM Date & Time Blend Conc 1/12/9 1:48 AM 1/12/9 12: PM 1/12/9 1:12 PM 1/12/9 2:24 PM 1/12/9 3:36 PM Strong [H2O2] decay observed over roughly 12 hours Similar effects occur in global loops, though possibly with different time constants depending on design 17

18 Fishbone Diagram Partially completed fishbone for particle qual failures GROUPINGS OF POSSIBLE ROOT CAUSES FAILURE Human Machine etc. Inadvertent change Wrong slurry/chemical Data entry error Wrong recipe Valve or pump fail Peristaltic tubing Leaks Calibration drift Control system Materials Methods Contamination End of filter life Slurry pot life Pad / conditioner life Slurry lot Poor process optimization Poorly designed loop PM frequency or scope Insufficient monitors Inadequate tool clean Things that make you go HMMM 18

19 Summary Slurry is one of the most critical ingredients for maintaining a consistent CMP process When excursions occur (and they do), the key is to find the problem quickly Follow a systematic troubleshooting approach Design (or redesign) slurry delivery methods to minimize risks AND accumulate the proper data for efficient troubleshooting 19

20 Thank you Contact Info: Robert L. Rhoades, Ph.D. Entrepix, Inc. Tel: Brian Orzechowski and Jeff Wilmer Divind, LLC Tel: