Anaheim, CA. Craig Nishizaki, director of ATE development at NVIDIA, took advantage of the International Test Conference held here this week to deliver a talk titled “Leveraging Cross-Operational Test Data for Manufacturing Yield and DPPM/RMA Improvements.”
In the lunch presentation, sponsored by OptimalTest, he described the rollout by NVIDIA's silicon operations department of OptimalTest tools at subcontractors throughout Asia. “We use the tools to improve wafer sort yields and to leverage test data to improve manufacturing yields,” he said. The goal is to serve the video game market for which the company is famous as a provider of graphics chips. “We will continue to make the imaginary appear real,” he said, but added that the company also serves such as industrial design, medical research, and other areas that rely on visual computing.
The company, he said, was founded in 1993 and now finds its chips used in everything from superphones to supercars. Lately, he added, the company has expanded into cloud computing.
As a fabless company, he said, NVIDIA must manage the manufacture and and test of the chips that support the markets in which it participates. Test and product engineers, he said, had concluded that the company's existing yield solutions had become less effective in improving yields and quality—test data was spread across multiple tools, it had become tedious to combine and analyze data from different operations, and there was no easy way to use data from one operation (wafer acceptance test, for example) at another (such as final or functional test). The company wanted a convenient way to implement yield monitoring and improvement, yield analysis, and customer return analysis, or RMA.
“We see excessive delays in being able to analyze parameters across all operations,” he said. “It's difficult for engineers to become proficient in all the tools and take action across operations. We need faster access to complete data.” With traditional approaches, he said, data might arrive in hours or days or never.
A huge volume of data is generated daily, he said, and engineers need automated notification of yield excursions. “With OptimalTest we have faster access to accurate data to increase yield and productivity,” he said, thereby enabling common yield improvements across different groups.
Using WAT and CP analysis, he said, the engineers can closely monitor WAT parameters and determine how they affect yields. “We can see effects of test hardware like probe cards and get information in minutes instead of hours or days. We can discover probe sensitivity before it's too late.”
“We made the decision this year to consolidate all our test data under OT to support end-to-end analysis,” he said, noting that the OT tools offer a unified source for manufacturing test data, supporting cross operational analysis to improve yields and quality. The result is better RMA correlation across all operations and the ability to screen excessive drift or temperature dependencies. The OT tools, he said, enable novel approaches such as viewing final-test data in wafer-map format, opening up new possibilities for yield improvement.
He cited as an example burn-in-related drift for automotive products. ATE measurements are taken at pre-burn-in final test, the data loaded into the OT database, post burn-measurements are then taken, and data from the OT database permits determination of the delta drift. If the drift is to large, the part is flagged as a fail.
The use of the OT tools poses no impact on test time, he said, adding, “We feel the OT tools are extremely powerful.” He said NVIDIA has established an ongoing collaboration with OT, involving virtual test (involving the creation of custom parameters and combinations to gauge quality of part) and data feed forward (involving a preventive signature profile with escape prevention rules and a real-time bin switching infrastructure to create safety net to avoid future RMAs).
“Partnering with OptimalTest provides a unified source for manufacturing test data for real-time manufacturing decisions, enabling us to resolve manufacturing problems and errors faster,” he concluded. “More ideas are to come. This is only the tip of the iceberg.”