There were three keynotes to kick off CDNLive Boston. Tom Beckley gave the Cadence keynote, then Professor Duane Boning from MIT brought us up to date on photonics, and finally Jim Borowick from Medtronics told us how to design a pacemaker. It is funny how often you hardly hear about some topic and then suddenly it seems to be everywhere. ADAS, which both Tom and Duane talked about (especially the lidar), is one. Silicon photonics is another topic like that, something I knew very little about until the end of last year (see my post Silicon Photonics ) when Lumerical presented at CDNLive in San Jose (see my post Schematic-Driven Simulation and Layout of Complex Photonic ICs ) with an encore performance in Boston, and Tom touched on it, too. Then Jim was talking about embedding electronics in bodies and hearts just a week after the keynote at Hot Chips was all about direct brain interfaces (see my post Neural Engineering System Design yesterday). I guess we'll all soon have a solid-state lidar on our foreheads with direct brain interface. Tom Beckley: Still Wants a Flying Car In May, Tom gave the keynote at CDNLive in Munich where he told us about watching the Jetsons growing up and always wanted a flying car. But he will probably have to settle for a self-driving car. Today, he gave the keynote at CDNLive Boston, and at this point in automotive, six months is a long time. Arguably more progress has been made in deep learning (a key component of flying cars!) than in the rest of history. The automotive industry is going through a lot of changes due what I call The Triple Witching Hour of Automotive : people are going to give up driving their cars, people are going to give up internal combustion engines, and people are going to give up owning their cars. This means that a core competence will be electronics and software and lean development, and the defensive moat of good engines will go away. Tom lives in Pittsburg, which is perhaps the biggest hub for self-driving technology development. Uber has had 80 self-driving test vehicles for a year and a half. When you get an Uber there, you don't know if it will be a self-driving one (which also comes with an emergency driver and a software engineer). The lidar on the roof of the current generation of experimental vehicles costs $80K so is not going to be the technology to use. But it will go solid state, probably with a four-layer chip design, with steerable lasers, silicon photonics, electronics and, maybe, MEMS. Duane Boning, in the second keynote, told us some of how that is going to happen. If you read Breakfast Bytes regularly, I don't think I need to go over the Cadence product line in detail, but I will call out a few things Tom mentioned: Firstly, everyone's prediction that very few people would do advanced designs is not true and there are 160 different companies using Virtuoso for Advanced FinFET design. Two big product releases this year (not in Tom's business unit) were the Xcelium simulation environment and the new Pegasus physical verification system. The big one for Tom, which he'd hinted at in Munich without naming it, was Virtuoso System Design Platform, where you can edit a the package and board with the chip still open and all the parasitics, etc., updated in real time. And those car companies can make use of Tensilica's range of automotive-ready processors. Indeed, in somewhat of a coup, Cadence's neural net Cactus Net, which runs on the Tesnsilica Vision DSPs, holds the accuracy record on the German Traffic Sign Database Benchmark. It's Time to Learn about Silicon Photonics Professor Duane Boning of MIT almost pleaded with us that we need to get educated on silicon photonics because it is ready. It is: Interesting Convergent Designable There are lots of complex specialized things in silicon photonics, but the two fundamental ones are a modulator, that allows an electrical signal to modulate a laser and thus produce an optical signal that has information in it. And a detector, to do the same thing the other way around, turn those modulated light signals back into electrical signals. A light wave guide looks superficially like a wire but it has a number of differences. Firstly, wave guides can cross without interfering, it is not a "short". You can also put more than one frequency of light down the same waveguide and thus run an entire bus through a single waveguide. The optical signals are very sensitive to temperature, and, like a car, they go around tight corners more slowly that straighter ones. By way of an example, he had a chip from MIT using light for a lot of the communication. The important thing is that there were no changes made to the process to add photonics to the basic "silicon" (which, of course, involves a lot more materials). It contained 2 RISC-V processors and at the top had 11 modulators to turn electrical signals into modulated light and 11 detectors below. These formed the main memory interface and it could be routed around 20 meters of fiber. They were getting 550Gb/s down a single fiber. Once you have a library of basic "standard cells", you can build more complex silicon photonic SoCs. This is much more standard than MEMS. Here is a Cadence photonic set of cells that works in Virtuoso (no, you can't yet take a billion of these things and put them together in Innovus). Another area of research is using phase changes to build deep-learning neural networks. They have managed to use 56 intereferometers to do vowel recogntion. But it really is computing at the speed of light was the weights are all set, the light just goes straight through the filters. The big market so far has been datacenters. It is moving from long haul, to top of rack, to backplane, and soon onto boards. But there are some interesting areas that take advantage of waveguides in unique ways such as miomedical, where they can be sensitized to gases and molecules. Above is another big application. The phase of each emitter can be controled separately so it can be used for image generation, but the big driver is lidar, where this can be used as the emitter and another die used as the detector. Since it is steerable, it is no longer necessary to have a big unwieldy $50K lidar on the roof that the CEO at DAC a couple of years ago said "your wife will not like." Well, you'd probably lose your job for less right now, but the basic point is right, they are ugly and unaerodynamic. So the summary is that the markets are there, the technology is manufacturable, PDKs exist, the design challenges are manageable. Key takeaways: Shared wafer processing, equipment, materials as normal "silicon" Refractive index (for waveguides) is good with high contrast Si in SiO2, Si3N4 in SiO2 (standard process) Can make small um scale bends Electro-optic Si devices made with ion implanting Integrated photodectors with germanium growth How to Design a Pacemaker Err...you probably don't want to try, after what Jim Borowick of Medtronic told us. For a start, you will be competing with them and they have 88,000 employees, are present in 160 countries, and at any given time are running over 400 clinical trials. We got a tutorial on how signals in hearts move around (and what happens when they don't, which is where Medtronic pacemakers come in). Here are a few more challenges you will face: They are implantable systems in a highly regulated market (by the FDA in the US) They communicate wirelessly with high security They have to provide high voltage (think heart defibrillators on a smaller scale) The battery has to last 9-15 years (and you don't get to change the battery over the life of the device) The chips are not especially large but I'm guessing there is hairy sub-threshold analog design to get the power so low, and obviously the reliability requirements are high. They use Palladium and Protium to develop the software. They use what I think of as the standard approach, using Palladium when the design is unstable and you need full visibility, then switch to Protium when software performance becomes more important. Post manufacture, they obviously run a lot of tests on the device, in real time, and then clinical trials. It's not enough that the chip works, it has to match the clinical needs of patients and their doctors. They have been taking a look at what they can do in the future with the technology. They took everything they know about system efficiency, shrink it, and put it in a pacemaker so small that it is a 1/10 of the size of a normal pacemaker and can be implanted directly in the heart up the femoral artery. It will last nine years. One challenge is to make sure it cannot dislodge since...let's say that would be bad. He had one with him and let me photograph it. I'm not sure if this is truly on the market yet or is in trials. I guess if you are in the market for a pacemaker, you can find out. It's called Micra. Oh, and if you think your design environments are cool, try having one with an animated heart Sign up for Sunday Brunch, the weekly Breakfast Bytes email.![]()
↧