ITF Keynote: IC Innovation—the Heartbeat of Yesterday, Today, Tomorrow

Luc van den Hove, the CEO of imec, gave the opening keynote at the imec Technology Forum in Brussels last week. There were really two themes to his presentation.

One theme was Gordon Moore and Moore's Law. imec had awarded Gordon Moore the Lifetime of Innovation Award. A couple of imec people had flown to Moore's home in Hawaii to interview him. The full interview will appear in the June issue of imec magazine, but at ITF they had video of some of the interview. Moore was asked who his own technology heroes were and he picked Bob Noyce, Andy Grove, and John Bardeen. When he was asked about what would be important in the future, Gordon Moore ducked the question, saying, "I missed the PC, I missed the importance of the Internet. Predicting is a tough job and I'll leave that to someone else." Moore's Law itself was a kind of prediction, of course, and it has been the heartbeat of the industry for 50 years. Back in that 1965 article, he also predicted:

Integrated circuits will lead to such wonders as home computers…automatic controls for automobiles, and personal portable communications equipment

Given than an integrated circuit held just a few dozen transistors in those days, this was an amazing prediction. We don't quite have self-driving cars but they seem imminent, and all of us carry around our personal portable communications equipment.

The second theme was that Moore's Law, in the sense of two-dimensional scaling, is coming to an end and that has implications both for semiconductor technology and, especially, the need to optimize entire systems, not just semiconductor manufacturing. However, Luc laid out what now looks to be the future of scaling: FinFETs today, leading to lateral nanowires (gate-all-round), and then vertical nanowires, taking us to the 3nm era. The big challenge is probably not technical, as long as EUV comes along, but economic. It is not good building wonderfully dense, fast, low-power transistors if they are prohibitively expensive.

But 2D scaling will eventually slow down and we need to go into the 3rd dimension. This is already happening for NAND Flash, but there are possibilities for logic to stack the transistors, especially for FPGAs which have a lot of regularity. This can be combined with heterogeneous integration using through-silicon-vias (TSVs), adding memory to the logic die and, perhaps, optimizing whole systems onto multiple layers that are then stacked during assembly.

But innovation needs to take place at higher levels, at the computer architecture level. One area that is increasingly important is neuromorphic computing. Neural nets of one form and another form the basis of most artificial intelligence (such as the system that beat the world go champion in Korea recently) and for a lot of the recognition tasks needed for autonomous vehicles.

The cloud will not be the only brain we need, Luc pointed out. We will need distributed brains since computation is cheaper than bandwidth. "Transmit wisdom, not the data itself", which requires local computing resource. This technology needs to scale up to entire cities. Imec is involved with the city of things project in Antwerp, which has 200,000 users, 100 gateways, 34,000 IoT devices. The core technology underlying all of these applications is semiconductor technology, where imec has world leadership. They are now combining that with iMinds, a combination research center and startup incubator, which is now being merged into imec.

Another important area is improving sensors for smart sensor systems. Sensors need to be unobtrusive, ultra-small, cheap, intelligent and energy efficient. One area where imec has been working is improving radar, an important part of making autonomous vehicles cheap enough to become a reality. As the frequency goes up, the size of the antenna required goes down and so it become feasible to put it on chip. imec had a demonstration later in the day of a 70MHz radar implemented on a standard digital foundry process. The next generation will be 140MHz.

Going up another level get to system-technology co-innovation, an acknowledgement that for an IoT application, the whole system from cloud-based software, to embedded software, computer architecture and the semiconductor substrate itself all need to be optimized together. imec can't do it all on their own, this requires an entire ecosystem. At the bottom are the semiconductor manufacturers: Intel, Samsung, TSMC, GlobalFoundries, and the memory manufacturers (Micron, SK hynix, Toshiba and SanDisk). The next layer are the fabless companies such as Qualcomm, Xilinx, Sony ,and Huawei. Then the application partners, too numerous to count, varying from equipment manufacturers, EDA and IP suppliers like Cadence, chemical companies, end application companies like Oracle and Google.

Next: DAC: My Guide

Previous: Inside Secure Writes the Book on IoT Security for Dummies