Optimizing product cost through modern test management
05/01/2007
Leading-edge fabs produce thousands of part numbers with new kinds of problems related to shrinking geometries, such as random defects from particulates and systematic fails caused by process alignment errors. Shrinking geometries associated with new process technologies have also introduced new test challenges-as a result, testing semiconductors today requires a sophistication only imagined by a few visionaries just five years ago.
The impact of these shrink-related issues-for example, much lower fab yields-has increased dramatically because competition demands that these products go to market shortly after first hardware, while also requiring improved performance, price points attainable only at high yields, and low-cost test with mission critical quality
eliability. These requirements can be met only by elevating test from a “backend” operation to a central hub with enterprise-wide, real-time processing capability. To that end, test management needs to concentrate on:
- Running an efficient test operation that may be multi-company, spanning the globe 24×7;
- Lowering all branches of the cost-of-test tree;
- Simplifying test development where manpower is at a premium;
- Obeying important customization rules;
- Enabling yield-and-reliability learning at a rate achievable only through real-time data sharing, analysis, and decision-making; and
- Seamlessly integrating all of the above within the existing infrastructure and business processes.
A modern approach to test management has several benefits. It can guarantee accurate, repeatable, and reliable test processes that will reduce both test escapes (i.e., shipment of bad product) and good product rejection. Also, instilling discipline through automation reduces set-up times, retesting, and the number of product lots on hold. When test time reduction (TTR) techniques, improved yields, and efficient test floors are used, capital expenditures are significantly reduced. One technique uses adaptive test so each specific device is subjected only to the amount of test needed to guarantee whether it is good or bad.
Test development can be optimized through menu-driven rule sets that allow easy generation and simulation of the rules required to test, sort, and disposition the product. This allows for real-time customization of test programs and chip-by-chip diagnostics. Additional offline software tools must be available for data mining and stacking of data from various process/test steps, enabling post-test analysis to identify trends and outliers. These techniques are critical for faster yield
eliability learning by utilizing the two-way exchange of data from fab processes, such as lithography through in-line tests and all subsequent test insertions. This analysis must be comprehensive to do an effective comparison on a die-to-die, wafer-to-wafer, and lot-to-lot basis.
As a result of these methods, what initially appears as a random fault can actually be pinpointed as a systematic failure. As Qualcomm’s Michael Campbell indicated at ITC 2006 [1], “Some systematic fails may require stacking data from many wafers to identify them. It’s a 3D puzzle.” The test management solution should fundamentally provide closed-loop learning. By using data from fab and test in conjunction with rules/diagnostics, the failure analysis tools can pinpoint root cause failure for corrective action in the fab. By having a highly automated closed-loop process, yield improvements are achieved at a much faster pace. At the same time, analysis of trends/outliers can be complemented with statistical methods to post-process test data for reliability prediction and burn-in reduction. Reliability learning is also accelerated.
IDMs perform bits and pieces of test management. Most fabless companies needlessly allow the foundry/subcon model to get in the way. In all cases, semiconductor companies are lowering profitability if they don’t move to comprehensive test management. Those that worry about the cost of implementing it need to take a closer look at the benefits of lower cost-of-test and much faster yield
eliability learning. The ROI is there for those who have the vision.
Reference
1. “Panel 7: Role of Test in Yield-learning for 65nm and Beyond,” 2006 International Test Conference, Santa Clara, CA.
 |
Contact John Bearden, director of business development, at OptimalTest, 20764 Tisbury Lane, North Fort Myers, FL 33917; ph 239/997-2613, e-mail [email protected]. OptimalTest is headquartered in Nes-Ziona, Israel.