Rattner on memristors, memory, power, raytracing and chip fabrication
Justin Rattner held court after his keynote in a room full of journalists, and answered a barrage of questions. Here are my notes from that meeting: I've paraphrased what he said, so don't take these as direct quotes, but they accurately reflect his answers.
On the problems of powering portable devices:
Memory interfaces today are dreadful. They consume way too much power. There are one or two orders of magnitude difference between on and off chip memory access. And it's terribly inefficient, you access thousands of bits and you only need to check one or two. We have developed some early thoughts you'll hear more about: we were hoping for the Solid State Circuits conference in February next year, but it'll be later now. If you operate close to the transistor threshold – a few tenths of a volt above – you can get eight or nine times the energy efficiency. Can you build a whole chip that way? We haven't built a processor that works that way yet.
We decided a while ago not to get into battery technology, but we are looking at nanotechnology applied to ultracapacitors – in next few years we'll be able to talk about that work, we think there's an opportunity to build extremely dense capacitive devices, build something that's purely capacitive and can harvest energy,and that can power devices.
On memristors:
Memristors – we've taken a pretty close look at them, and are involved in any number of collaborations with HP Labs. They're one of a whole collection of device possibilities for non-volatile storage. People have been asking for a while - what's beyond Nand flash? We thought we were on the edge with 50nm, and people were predicting the end – but now down to around 22nm, so Nand flash broke through that barrier. Memristors, we're not convinced that they offer any particular advantages over the alternatives. It's one thing to show the device in the lab, another to make it.
We've shown large phase change memories, so the alternatives to memristors are much further down the road,. HP has good stuff in the lab, but it's a long way from product.
On raytracing graphics, and Intel's fondness thereof:
Raytracing – we're big believers in the technology. We were hoping to see it sooner in games, but gamers are concentrating on low density graphics so the advantage of raytracing isn't nearly as great. Where you will begin to see it in the next year or so is in the 3D web, where you want to create your own visual content but don't have the time to spend doing the lighting and tracing and things like that. Raytracing is good match to that, you can render photorealistic images without being a graphics guru, you get the right result without much effort.
On the future of lithography in chip fabrication:
We made substantial investment in computational lithography. Some of world's biggest supercomputers are in our fabs. You take the masks and predistort them mathematically to correct for aberrations in the optical path. As that moves forward, people move from single masks to multimask patterning as they try to extend the life of optical. X-ray or extreme ultraviolet (EUV) are still the holy grail, but are much more difficult and much much more costly than expected. The boundary, where it's more effective to move to the new technology instead of adding more masks, keeps moving out.