CAST Workshop Explores Barriers to Expanding Adaptive Test
Test industry experts from chip manufacturers, electronic system manufacturers, ATE suppliers, and Adaptive Test software providers gathered to discuss the opportunities, challenges and barriers to successful and cost-effective Adaptive Test implementation.
Adaptive Test (AT) has been defined as a "broad term used to describe methods that change test conditions, test flow, test content and test limits (potentially at the die/unit or sub-die level) based on manufacturing test data and statistical data analysis.” The International Technology Roadmap for Semiconductors (ITRS) concludes that AT methods will deliver benefits including "lower test costs, higher yields, better quality & reliability and an improved data collection environment for more effective and rapid yield learning."
Some IDMs have implemented various AT methods, but many fabless semiconductor companies and others with multiple products and test flows have struggled with implementation and with isolating the cost savings to verify the AT business payback. Without common definitions, interfaces and protocols, expectations from customers in the value chain (including OSATS and system OEMs), and acknowledged Best Practices, many believe there is a need for industry-wide collaboration, guidelines and standards to enable cost-effective implementation of AT methods.
Caption: Adaptive Test Flow/Architecture (Source: Phil Nigh, IBM, Chairman, Adaptive Test Subgroup, Test & Testability Section of the ITRS)
AT is typically applying feedforward data derived from ATE and handlers, including die ID, parametric data, timing, IDDQ, device failures and other data, for real-time analysis and optimization. Statistical analysis of this real-time data can then be used to dynamically change test flows, limits and parameters in the ATE, potentially improving yield, quality, reliability and cost reduction. Post-test analysis and dispositioning data can also be fed back and forward into the test flow to change statistical bin limits to more meaningful levels. In practice, both real-time and post-test data can deliver significant insights and improvements on yield, test time and failure rates.
The workshop began with a panel of speakers currently deploying AT in their test operations, including Phil Nigh from IBM and chairman of the ITRS subgroup on Test and Testability, Glenn Plowman of Qualcomm, Jeffrey Roehr of Texas Instruments, and Matthias Kamm providing the customer perspective from Cisco.
Phil Nigh worked with a cross-section of companies on ITRS, and he described various implementations of AT — including real-time analysis and sampling for test pattern reduction to reduce test times. Nigh explained that IBM has performed a couple of different types of “test pattern sampling” methods: at wafer probe to select the optimal test pattern set for specific wafer based on sample probing results, and at final test where they use historical data and off-line analysis to drive real-time sampling and dynamic analysis. IBM also uses data feedforward using chip ID, which was originally purposed to feed-forward RAM repair data, then deployed for many other useful purposes, such as parametric data from one test step for use in future search/limit optimization.
Plowman provided an overview of how Qualcomm deploys AT at its test subcontractors to use real-time and near real-time statistical process control on data from in-line test. This data is used to adjust test limits and content “on-the-fly,” thereby improving yield and reducing test time and cost. He provided insights and recommendations into the process of rule building, simulation, tester integration, and data logging architecture.
Roehr provided practical advice on how to develop and deploy AT in the real world, including the typical need to provide basic statistical training for test engineers and the need to collect sample test run data for statistical analysis. From there, it’s a matter of determining how to write AT tests, involving design teams into Adaptive Test solutions, and how to integrate production planners so that they understand how to react to AT yield-related loss. For AT to be successful, a 100% daily exposure to statistical AT elements is required throughout the entire design to production test chain. To be successful, “Adaptive Test must become as universal as traffic lights.”
Roehr described how successful AT implementation requires “data, lots of data.” Every wafer and IC chip, from cradle to grave, needs traceability; including wafer inspection and parametric data, probe data per die, and assembly information. Test and burn-in data on multiple passes are required to fully implement AT. To achieve the ITRS vision, board level test data and customer use (failure) data are also required.
Adaptive Test at the Card/System Level
Mathias Kamm of Cisco explained how his firm applies AT to use data that is fed forward or backward to enhance test applied to the card/system — inside and outside of the plant. From Cisco’s perspective, data must be carefully structured to make it available throughout the manufacturing process. Many data attributes are evaluated including die identification, location on wafer, latency, persistence, data size, security, and other variables. The database architecture consists of local DB for real-time ATE/test cell operation where latency is 1-2 seconds and retention is hours, non-real-time DBs for lot size and dispositioning decisions (latency in minutes; retention for days), and company-wide DBs for long-term storage.
For card level AT, the challenges include dealing with more functional and fewer parametric tests (Iddq, Fmax, etc.), a potentially high mix of various component types, product specific tests and diagnosis, variability of traffic tests for each product, and the lack of standard 3rd party offerings. Traceability is a key requirement: each component is traced by lot code, date code, unique board serial number and ECID (100 percent on ASICs). The benefits include tracked ASIC repair flow at the repair station, optimized test flow (less rework, lower cost, and avoidance of mass recalls).
Other card level applications for AT at Cisco include analyzing CND (cannot duplicate fault) treatment/routing, card-tailored test based rework history, potential tailored tests, test based on card maturity, intelligent sampling and customized test that match “fail symptoms”.
After the opening presentations, the core of the workshop was an interactive discussion of key challenges and opportunities for industry-wide AT adoption. In addition to chipmakers, it included technical experts from ATE and Adaptive Test software suppliers. The wide ranging discussion included the barriers to industry-wide die level identification, industry standard algorithms, lack of statistical knowledge in the test flow, and the difficulties in achieving OSAT/subcon support. The industry still lacks standard definitions and terminology and ATE interfaces and data formats vary widely.
Among the action items was a decision by ATE suppliers to explore and discuss standard data interfaces. An industry standard data interface to test equipment, while believed not to be a useful means of differentiating ATE equipment, would significantly reduce implementation time and cost for AT and ATE suppliers, chip companies and test houses. It would also facilitate more widespread usage of common rules, algorithms and process control/statistics with the knowledge necessary to implement comprehensive AT strategies.
CAST and the ITRS subgroup on Test and Testability will continue to explore opportunities for industry-wide collaboration and education to realize and benefit from industry-wide adaptive test implementation. For more information on the ITRS efforts in Adaptive Test, visit http://icdt.ece.pdx.edu/~icdt/cgi-bin/adaptive.cgi/AdaptiveTest
For more information on CAST, visit www.semi.org/cast
April 3, 2012