Executive Editor
The semiconductor industry is facing severe challenges. Moore’s Law, encountering economic pressures and physical limits, will not continue much longer. With the end of advances in traditional scaling, the industry will need to look to alternatives to meet tomorrow’s performance needs.
There is potential good news for test engineers. Wally Rhines, Mentor Graphics CEO, delivered a keynote address at the recent 2016 International Test Conference in Fort Worth in which he said that chipmakers have competed on features like price, performance, and power—not on how much they test their chips. Consequently, test has been considered overhead. That is changing as it becomes necessary for foundries and their fabless customers, for instance, to share data. “The value of test data will increase,” he said. “Test as overhead becomes test as value creator.”
Semiconductor makers are aware of this. For example, Roberto Lissoni, ST corporate quality director, delivered a talk at ITC in which he described working with Optimal+ to deploy test as a strategic asset to increase company performance. (His talk was the latest in a series of ITC sessions hosted by Optimal+.) Lissoni said big data is the new wave of structural change in the semiconductor industry, describing ST as a “brain-on data-centric company” merging expertise in semiconductors, computer science, and math and statistics.
The challenges will only become more daunting. In another ITC keynote address, Rob A. Rutenbar, Bliss Professor and Head, Department of Computer Science, University of Illinois at Urbana-Champaign, stated, “The unfortunate reality for today’s fab now is that every chip behavior is a little smear of probability.” And in yet another address, Ken Hansen, CEO, Semiconductor Research, alluded to challenges of big data itself. He said he’s heard people refer to data as the new oil, but he prefers to consider it the new gold—you have to sift through a lot to find a valuable nugget.
There are several ways forward for the semiconductor industry. Rutenbar said that as software-centric approaches that rely on Moore’s Law lose their viability, it’s advisable to “throw transistors at the problem” through the use of hardware accelerators. He described a belief-propagation inference method that shows promise for resilience when implemented on a stochastic fabric.
Heterogeneous integration also is a way forward, according to Dave Armstrong, director of business development at Advantest, who introduced a technical session on the topic at ITC. “To me, the end of Moore’s Law is certain,” Armstrong said. “Heterogeneous integration is poised to drive the industry forward.” Armstrong introduced the Heterogeneous Integration Roadmap (HIR), saying, “Roadmapping is an incredible way to put the pieces together” by gathering experts in one forum.
The HIR team kicked off earlier this year with 576 participants representing more than 100 companies. Armstrong said the HIR test working group includes more than 70 professionals representing more than 45 companies. The test group has 13 subcommittees, focusing on topics including yield learning, RF, probe cards, test sockets, adaptive test, high-speed I/O, memory, analog/mixed-signal, DFT, handlers, adaptive test, and specialty devices like MEMS.
You can participate in the move forward. Throughout his ITC presentation, Armstrong urged volunteers to support the HIR effort—visit my blog for more. Hansen called for multidisciplinary participation within the context of his organization, noting that interesting things happen at the boundaries of chemistry, physics, computer science, mechanical engineering, and electrical engineering.
Hansen noted that the test problem has been considered “NP hard” and will continue to be. Test engineers, he said, will need to contend with stochastic devices, new architectures, new package-integration technologies, security issues, and Shannon- and brain-inspired techniques. He concluded, “There is more and more excitement for test than there has ever been.”
