Test Moving Forward And Backward

Test, once considered an important but rather mundane way of separating good chips from the not-so-good and the total rejects, is taking on a whole new life.

After decades of largely living in the shadows behind design and advancements in materials and lithography, test has quietly shifted into a much more critical and more public role. But it has taken several rather significant shifts across the industry to make this happen.

First of all, the customer base has changed. Markets have fragmented, and companies buying or building chips for those market segments are looking for increasingly customized solutions. Rather than settle for 20% to 30% improvements in power and performance afforded by shrinking features, they are designing (or helping to design) custom architectures that are extremely fast and much more power efficient. These are not one-size-fits-all types of chips, and often they are using some version of advanced packaging and different types of processing elements and memories.

What’s particularly noteworthy is the number of respins, which are on the rise everywhere. First-time silicon is no longer the mantra, and this has opened the door for test to play a significant and entirely new role. For vertically integrated companies such as Apple, Cisco, Google, HPE, IBM, Facebook, Amazon, Alibaba and Baidu, among others, one or two respins are an acceptable cost of doing business. That cost is amortized across an overall system budget, and there is enough padding to pay for iterative improvements in chips. Test is becoming a critical piece in this process because results of testing are being fed back into the design chain to fix problems before the chips are used commercially.

What has made test data especially valuable is the application of AI and machine learning to those processes. Whether it’s done by the test companies themselves, or whether analytics are applied to test data after the fact, this represents an important shift in the value of that data earlier in the design-through-manufacturing flow. Being able to spot patterns at the test phase and then make sense of them is a major shift, and it will become increasingly important as chips replace mechanical functions in automotive, medical and industrial applications.

A third shift involves an increasing reliance on in-circuit monitoring, which has become a mainstay in designs at 10/7nm and below, with a growing presence in automotive applications. What makes this so important is the ability to spot irregularities after these chips have been in use in the market for months or years. This is important from a reliability and liability standpoint, because if the electronics in a steering or braking system are about to die, or if a mission-critical server is about to overheat, it’s important to first address the immediate issue, and then to prevent it in future generations of chips.

In-circuit monitoring and testing has been around for a while, but it is rapidly gaining acceptance across a variety of markets. As it does, the data generated from this kind of real-time monitoring is being fed back into the design through manufacturing chain, where it is being analyzed with increasingly sophisticated tools to determine whether failures are truly random or whether there is a pattern in a certain batch of manufactured parts.

Taken as a whole, this puts test and test data squarely in the center of the entire supply chain for semiconductors. And for the first time in many years, test is now considered a vital part of architectures, manufacturing plans, and post-production analytics. What a difference a couple of years make.