4 Issues In Test

When most design engineers think about test, they envision a large piece of equipment in the fab they probably will never actually see or interact with. But as chips become more complex—driven by an explosion in both quantity and different types of data—test is emerging as one of the big challenges in design and manufacturing.

There are four primary segments for test, each with its own set of requirements and issues, many of which are recent. On the front end are lab testing and built-in self-test. Lab testing is so closely tied to chip design that most people don’t think of it as test. In fact, some of the techniques and equipment used in the lab pre-date the integrated circuit. In years past, when televisions relied on vacuum tubes, TV repairmen would attach the same kind of oscilloscope used in a lab to measure the amplitude of an analog signal and figure out which part had gone bad. Oscilloscopes are still being used in mixed-signal labs, but they’re now complemented by a whole raft of equipment to measure power draw, signal integrity, noise, and usually connected to a virtual dashboard that can run various tests from a workstation.

The problem on the design side is that data from the lab can’t necessarily be used anywhere else because it’s in a different format. There is work underway to try to bridge those two worlds…finally. This was a back-burner issue for many years. What makes it very compelling to solve these days is an increase in the volume of tests that need to be run.

Understanding the behavior of heterogeneous systems, particularly if there is some element of AI or machine learning in there, is an ongoing process because these devices evolve over time. Because there’s currently no visibility into how these systems change, as much data needs to be collected from all parts of the design through manufacturing flow to be able to piece together what caused outlier defects that are beyond an acceptable distribution of behavior

Along with this, there needs to be much more built-in self-test to understand how circuits are behaving in real time. Some of this can be done in real-time using in-circuit monitoring, which is what companies such as Moortec and UltraSoC are doing today, but some of it requires standardized testing. The problem is that as more features are packed into these devices, more test circuitry is required. In some cases, this could amount to more than a third of the real estate on a die, and it could impact the overall performance and power budgets for a device. Still, in an automotive or industrial application, where safety is part of the picture, BiST is a known solution. The question now is how to add efficiencies in terms of power, performance and area, potentially even through concurrent testing.

A third problem area is on the ATE side, where more complicated and denser circuitry—and more possible interactions with other devices inside or outside of a system—requires multiple tests to be done at the same time, including some in the context of a working system. While this may seem like a fab test issue, it’s also a design-for-test challenge. If a design team doesn’t get this right up front, testing could miss problems at the manufacturing stage, leading to low yield, chips that don’t work, or arguably even worse, chips that prematurely fail in the market. The cost of a recall of a car is far greater than fixing the problem before the vehicle leaves the factory.

A fourth challenge involves new structures and packaging that make it impossible to use existing testing approaches. The normal method of attaching a lead to a tester and running a variety of tests for such things as temperature and power no longer work. In the case of 5G, which will be a critical component of automotive safety in places like China, where the emphasis is on centralized data processing and mining, this will require over-the-air testing. So far, this is more theory than reality. The same could apply to some types of advanced packaging, where really understanding what is going on inside a 3D-IC, for example, is difficult to know from the outside.

For a sector that used be be the rather staid and stable workhorse of semiconductor manufacturing, test is no longer something that can just be ignored. It’s now an integral and essential part of heterogeneous and homogeneous system design all the way through to manufacturing and even beyond. And don’t be surprised if costs begin to rise, because solving complex problems that span the breadth of interrelated processes for increasingly complex designs isn’t going to be cheap.