Issue link: https://iconnect007.uberflip.com/i/1295812
62 DESIGN007 MAGAZINE I OCTOBER 2020 just because you need it in one key section has always been onerous. This does give you the opportunity to flex with what's inside the package. Demmin: The most obvious example of that, which has been recognized for a long time, is memory and processors. They use differ- ent fabrication processes. It's still silicon, digi- tal, etc., but they're different processes just because of the nature of how they function. Splitting memory off from processors has been common for quite a while. But all the other functions that end up in a big system-on-chip piece of silicon makes it quite inefficient to do it that way. Designers do everything they can to minimize the area's silicon usage. And if you can carve out some- thing that doesn't need that leading node, that's a smart thing to do, but you need to put it all back together after you carve it up. That's the heterogeneous integration angle need that arises from this partitioning. Johnson: Your target audience and the indus- try you're writing about have a reputation for being fairly conservative. What would be the motivation for the mil-aero industry to jump into this technology? Demmin: It's different depending on what indus- try you're in. One of the challenges is that the HIR effort covers everything from consumer to military. Companies that are on the digital high-volume side that are very supportive of this—such as Intel, AMD, and NVIDIA—can save a lot of money and create new capabilities if they can divide up their functions and then integrate them after the optimized manufactur- ing of each. Xilinx was the first to divide up one of their big FPGAs, and they split it into four slices and put them on a silicon interposer that con- nected them together. Nowadays, that's pretty common to do something like that; compa- nies like Intel are integrating different kinds of devices—not just the same device divided into more manageable chips. There's a lot of inter- est in heterogeneous integration of processors, data converters, memory, ASICs, and even optical components now. On the military side, in spite of what people might envision as unlimited funding for mili- tary stuff, that's not really the case—especially when the volumes are low, and you don't have the total volume driver as you do in cellphones to amortize design, mask, and fab costs. Hav- ing those same kinds of savings, while still mixing and matching the best types of devices for different functions, offers that benefit for the aerospace and defense world. It has even more challenging because there would commonly be a wider variety of devices potentially in a radar system or defense com- munication or electronic warfare where you need to be able to put together silicon, gal- lium arsenide, indium phosphide, and gallium nitride, among other device types to get the best possible performance. Challenges are con- ceptually similar in military and aerospace, but there are typically more types of devices for these specialized applications. Johnson: What was your team's process for putting together the chapter? Jeff Demmin