Statistical Quality Control of Wafer Level DC Die Sort Test Y.Z. Wang, R.S. Persaud, R.C. Salvador, D. J. Troy ANADIGICS, INC. 141 Mt. Bethel Road, Warren, NJ 07059 Tel: (908) 668-5000, e-mail: ywang@anadigics.com Keywords: Die Sort, statistical quality control, yield, yield excursion Abstract In this report, we developed a statistical quality control (SQC) system associated in our test programs to control wafer level DC Die Sort test by providing controls over TRI failures, bin yield loss and mean and standard deviation variances of key parameters. This SQC system has dramatically reduced the test caused yield loss ~1.2% since the tool was implemented. The yield excursions have been captured in real-time from the control of bin yields and variability of the critical parameters. Productivity is also greatly improved by integrating automated recipe downloading and TRI re-test as well as Die Sort disposition decision making into the process. INTRODUCTION The increasing competition and the challenges of cost reduction have forced the GaAs manufacturers to pursue the improvements in die yield and cycle time [1]. The ANADIGICS wafer fab is working towards a goal of a “Good Die Fab” mentality (as opposed to a “Good Wafer Fab” mentality) [2]. To achieve this goal, we would not only need to maintain higher cumulative wafer yields, but it must also increase die yields. This requires us to reduce the yield loss at the wafer level DC Die Sort test to ensure die level compliance. The Die Sort test is used to screen out die level defects, process or material variations resulted in device or circuit failures as measured on the die [3]. The criteria that we had used to disposition products at Die Sort for some time was to apply a minimum yield limit for each wafer by product. However, a large quantity of die loss may be incurred due to test-related failures. Since the measured Die Sort yield was above the minimum yield standard, the wafer would continue processing and the learning regarding the failure would be lost. In addition, the wafers showing unique process or material signatures may pass the minimum yield without requiring engineering attention. If the process or material caused electrical parameter shifts, we would not know until generating the historical trend charts. The objective of statistical quality control (SQC) at Die Sort test is to provide a control system by implementing controls over test related failures, Die Sort bin yield loss and critical parameter stability. This includes the integration of control limits which are imbedded into the Die Sort test program for each product while providing a system to perform an automated Die Sort disposition. With an SQC control system, we are able to reduce the Die Sort yield loss on testing related issue (TRI) caused failures, which in turn relieves engineering disposition resources while reducing the cycle time. In addition, yield excursion and process shifts are captured in real-time on wafer to wafer base. DIESORT TEST PROCEDURE ANADIGICS wafer level DC Die Sort test is performed in an integrated PXI test platform interfaced to an EG4090 prober. The tester delivers the measurement capability required by the production with high breakdown voltage capability, low-current measurement accuracy and flexibility [4]. The PXI-based tester provides a calibrated system that met the high-speed throughput as well as exceeding the acceptable gauge R&R requirements for key parameter measurements. This is a prerequisite condition for statistical quality control of Die Sort probing. Periodic golden wafer verification procedures have also been established to ensure the test gauge capability. For Die Sort testing prior to the SQC implementation, an operator would load the test program template for the product that was to be probed, and then loaded the first wafer, aligned the wafer and started to probe. When the whole lot was completed Die Sort test, the operator could reload the lot to test the failures if the Die Sort yield is below the minimum yield standard, or operator could move the lot to next process step when passing the minimum yield standard. For wafer passing the minimum yield standard, operator had no visibility whether the failed wafers were caused by TRI failures or by the unique material or process defects. Running a retest was completely dependent on the operator’s judgments or the engineering disposition requirements. In some cases, it even took a longer time for operator to reload the lot than to retest failures on wafer. Running a retest at Die Sort was a time-consuming process. In the SQC of Die Sort probing, we developed a system for in-situ control of the Die Sort test. The SQC system is associated in the product program template. It will be loaded when pulling out a corresponding test program by inputting a lot number. When a wafer is completely probed, the SQC disposition information of the wafer will show up on the CS MANTECH Conference, April 24-27, 2006, Vancouver, British Columbia, Canada 79 screen. Operator follows the guidance to take the actions. Figure 1 shows the SQC test flow chart. When a wafer needs to be retested, the operator will be guided to do so following the OCAP procedure without unloading the wafer. That greatly saves on test time. be close to the sanity limits. However, it is necessary to adjust the limits for some parameters so that the TRI failures are really captured. These types of failures are easily recovered by re-testing failures while adjusting either alignment, Z-height or by cleaning the probe tips. 14 Lot Start Yield Improvement from TRI, % 12 Load Wafer Probe Wafer Save "P" file & Track Disposition Message Delta Yield,% 10 8 6 4 2 TRI Failure? Yes w/TRI -Retest Message 0 Load Next Wafer 1 No w/ Other Messages 15 22 29 36 43 50 57 64 71 78 85 92 99 106 113 120 127 134 141 148 155 162 169 176 Wafer Number Continue Retest Unload Wafer Yes w/ TRI -on Hold message Figure 2. The delta yield improved from TRI re-test of a single product with the SQC. Follow OCAP? Yes OK No w/ other messages More wafers to test ? End Test 8 Retest/Continue Mode Select No Retest Failures TRI Failure? Save "R" File & Track Message Figure 1. The Die Sort test flow chart using the SQC. RESULTS AND DISCUSSIONS The Die Sort SQC system consists of three parts: TRI failure control, Die Sort bin control and parameter 3sigma control. For each key parameter the control limits are set up in these three parts based on the historical data. The SQC priority is in the order of TRI failure, Die Sort bin and then parameter 3sigma. The SQC results will show up in the screen and it is also saved in the Die Sort data file when the wafer is completely probed. Figure 2 shows the delta yield improvement from TRI retest of a single product with the SQC control. The improved yield is determined by the difference between the wafer yield with TRI re-test and the initial yield, which varies from near 0% up to >10%, and the average yield improvement is ~1.20% per wafer. From the improved yield in Figure 2 we can see some wafers have higher yield loss on the TRI failures, these failures are normally distributed in different parameters. Figure 3 shows the yield loss comparison of the electrical parameters before and after the implementation of SQC for the same product shown in Figure 2. This figure clearly shows the significant yield improvements for most parameters with the introduction of SQC. From Figures 2 and 3 we can see ~ 0% improvement for some wafers or some parameters. It indicates the catastrophic failures may be the real failures from circuit open or shorting caused by defects. The TRI limits cannot completely distinguish the TRI failures and the defect caused failures. In order to reduce unnecessary re-test, a limit for TRI failed die count limit is put in place for each parameter. 2. 1. TRI Failure Control From the Die Sort historical data analysis, the number one failure mode at Die Sort was the TRI caused yield loss prior to SQC implementation. Probing misalignment, probe contact and the contamination of probing tips were the possible root causes of the TRI failures. The TRI failures contributed the most to catastrophic failures recorded at Die Sort test. For most parameters the control limits of TRI can 80 Die Sort Bin Control The Die Sort test can screen out electrical failures that are outside the parameter specification limits and defect caused circuit or device failures. The defect limited yield (DLY) loss has a random variation from parameter to parameter and wafer to wafer. Whereas the parametric limited yield (PLY) loss may indicate the material or process is out of control or the design has deviated significantly from the target value on a given product. When implementing CS MANTECH Conference, April 24-27, 2006, Vancouver, British Columbia, Canada parameter yield loss control at Die Sort test, it is easy to capture the yield excursions. From the data analysis, we can detect root cause quickly. For a new product we are able to accelerate the yield learning, too. 0.3 Pre_SQC 0.25 Post_SQC Yield Loss, % 0.2 0.15 0.1 (b) 0.05 0 RB1_bias RB2_bias RB3_bias Ipd_Vm0 Imode_Vm0 VB2_Vm0 Imode_Vm3 IccSunLK Parameter Figure 3. The yield loss comparison of the electrical parameters pre and post the implementation of SQC. Figure 4 shows the pie charts of the limited yield loss comparison with and without SQC implemented for a single high running product. In the charts, the yield loss is split into TRI, DLY and PLY yield loss respectively. Prior to the SQC implementation the TRI yield loss was the highest in total yield loss. With the implementing of SQC the total yield loss is reduced, and the losses due to TRI are dramatically decreased from ~40% to ~16%. Both DLY and PLY losses are almost equally split for the remaining yield loss at Die Sort. Figure 4. The limited yield loss (a) without SQC implement, (b) with SQC implement. Figure 5 shows VB2_Vm0 yield excursions of a CDMA power amplifier product at Die Sort. In a certain time frame more than 5 production lots had VB2_Vm0 out of the bin control limit and it also caused higher yield loss. The yield loss was up to 8% per wafer. From the data analysis the VB2_Vm0 failures were found to be related to the NV1 openings where the first layer interconnections to resistor or base contacts are located. The SEM pictures on the contact revealed that a damaged stripe next to the contact resulting in VB2_Vm0 failures at Die Sort. It indicates that yield excursion can be captured and a problem can be isolated in time using bin control in place of the minimum yield loss. 9.0 VB2_Vm0 % 8.0 VB2_Vm0 Yield Loss, % 7.0 6.0 5.0 4.0 3.0 2.0 1.0 0.0 1 14 27 40 53 66 79 92 105 118 131 144 157 170 183 196 209 222 235 248 261 274 287 300 313 326 Wafert Number Figure 5. VB2_Vm0 yield excursions of a CDMA power amplifier product at Die Sort. 3. (a) Parameter 3sigma control In the third part of Die Sort SQC program, the across wafer averages and standard deviations of the critical parameters are monitored to determine the significant changes in the mean and in the variance. The control limits CS MANTECH Conference, April 24-27, 2006, Vancouver, British Columbia, Canada 81 are calculated using +/- 3 sigma of the parameter historical data to indicate the boundary limits for the mean and standard deviation. When the mean shifts or the variation changes, it implies that the process or material have shifted and requires investigation to return the parameter back under control. Figure 6 shows the bias resistance deviation control chart from a major product during a specific time frame. From the figure we can see the deviation of bias resistance is out of control starting from wafer No. 120 in the chart. This variation of the bias resistance deviation did not cause the bias resistance to go outside of specification limits, and there were no yield losses. However, the bias resistance showed a “bull’s eye” pattern at the wafer center. It indicated that our bias resistor process was out of control. From a fusing current test that is used to force device damage by sweeping current, we found that the resistors within the “bull’s eye” failed at the resistor to metal contact. This was different from the normal resistor, which had the resistor itself failed as a result of the fusing test. The root cause was identified as inadequate photo process latitude which resulted in the slight plasma damage to the contacts. The problem was subsequently fixed when the root cause was identified, and the process was back in control from wafer No. 300. This example demonstrates the importance of 3 sigma statistical control of the critical parameters at Die Sort. R_Bias Across Wafer Standard Deviation 14 12 CONCLUSIONS We developed a SQC system associated in our Die Sort test programs to control TRI failures, bin yield loss as well as stability of the critical parameters. This SQC system dramatically reduces the TRI caused yield loss at Die Sort test, and overall ~1.2% yield is improved with the auto TRI re-test since the tool was implemented. The yield excursions caused by the process or material variations have been captured in real-time from the control of bin yield and mean and deviation of critical parameters. Productivity is greatly improved by integrating automated re-test and Die Sort disposition decision making into the process. The methodology is so successful that it has been propagated to our PCM testing process. ACKNOWLEDGEMENTS The authors would like to thank Mr. Henry Ang of ANADIGICS for his contributions to this work. We would also like to express our appreciations to Mr. Gregory Guth of ANADIGICS for the useful discussions. REFERENCES [1]. J. Bordelon, et al., IEEE Design & Test of Computers, Vol. 19, No. 3, May 2002. 10 8 [2]. R. B. Miller, W. C. Riordan, Proceedings of 2001 International Test Conference, pp. 1118-1127, Oct. 2001. UCL 6 4 UCL [3]. P. Bohlinger, Y. Z. Wang, T. McGuine and C. Nielson, 2001 GaAs MANTECH Technical Digest, pp. 37-40, May 2001. 2 LCL 514 495 476 457 438 419 400 381 362 343 324 305 286 267 248 229 210 191 172 153 96 134 115 77 58 39 1 20 0 Wafer Number Figure 6. The bias resistance standard deviation SQC control chart from a major product during a specific time frame. The wafer level Die Sort testing was a bottleneck to throughput because many lots need engineering attention before continuing to process. Since the SQC of Die Sort test was implemented, the TRI failures are going to be re-tested immediately following the wafer Die Sort test. The automated disposition information for each wafer is clearly shown in the Die Sort map. Operators can move lots following the disposition message without the engineer’s involvement. From the comparison pre and post SQC implementation, the re-testing rates are comparable. 82 However, the wafers held for engineering disposition were dramatically reduced ~65% by implementing SQC. With the application of SQC, wafer level DC Die Sort test can also use sampling strategy when the 100% automated defect inspection is applied in wafer fab. All these efforts greatly improve the productivity of Die Sort test and reduce the overall cycle time. [4]. Pickering Interfaces, User Manual of Matrix Module, Issue 3, pp.6.1-6.2, March 2002. ACRONYMS SQC: Statistical Quality Control TRI: Testing Related Issue DLY: Defect Limited Yield PLY: Parametric Limited Yield CDMA: Time Division Multiple Access OCAP: Out of Control Action Procedure PCM: Process Control Monitor CS MANTECH Conference, April 24-27, 2006, Vancouver, British Columbia, Canada