The new season of CDNLive kicked off earlier this week with CDNLive Silicon Valley in the Santa Clara Convention Center. The giveaway this year for visiting enough of our partners in the expo was the cute bear on the left. For those of you not from the area, in the background of the photo is Levi Stadium where the "San Francisco" 49ers play their home games. Like the New York teams, San Francisco no longer plays in the city that bears its name, but at least they play in the same state! There's even a Cadence angle. Levi Stadium was built on part of the parking lot for the Great America theme park. They have one of those rides where you get taken to the top of a tower and dropped called Drop Zone. But this ride replaced an earlier ride that was not as high, called The Edge. SDA, one of the forerunners of Cadence, had offices that looked out into the theme park. The marketing guy at the time told me that he named the original SDA products with "edge" names (like ChipEdge) because he was looking at the ride on the day he was coming up with names and didn't manage to think of anything better. That was over 30 years ago, before Cadence existed as a company. I'd like to say that there is a direct line from today's Innovus back to ChipEdge, but I'm not sure how true that is. However, SKILL was part of ChipEdge and continues to live on. Here's a paragraph from a 1986 Dataquest report (apparently we didn't yet have decent word-processing or laser printers): Keynotes Anyway, that's enough about the distant past. In the very recent past, just a couple of days ago, Lip-Bu Tan, Cadence's CEO kicked off the keynote for CDNLive. In some ways, semiconductor has been a mature industry for the last few years, with modest growth and few startups (most of which, as was pointed out at the recent ESD Alliance CEO panel, Lip-Bu funded as part of his other job as Chairman of Walden International). That has changed. The growth last year was huge. Yes, some was the fact that DRAM is out of capacity and so memory prices strengthened. But even outside of memory the growth was solid. For the first year ever, the size of the semiconductor industry passed the $400B mark, and there is optimism in the whole ecosystem that we are on the way to a trillion dollar semiconductor industry within the next decade. There are also other venture capitalists funding fabless semiconductor companies, many in artificial intelligence, but by no means all. Lip-Bu called out five areas that are driving current growth. Mobile is the biggest market for semiconductor, although the growth has flattened from where it was a decade ago when smartphones exploded following the announcement of the iPhone. Automotive semiconductor is a segment in transition, with an older micro-controller-based market for simple controllers, and an advance market on leading-edge processes for ADAS and autonomous vehicles. Of course, that is a market that mostly remains in the future, but it is inherently big since there are about 100M vehicles sold each year. The next markets are edge computing and datacenters. These two go together since there is an expectation that there will be [fill in your own number of billions here] IoT edge devices sold, and almost all of them will also have some sort of backend that runs in the datacenter. But there is a lot of other stuff running in datacenters, and a portion of that is driven by the 5th market that Lip-Bu listed... Machine learning. This, again, is expected to be in two parts, with learning done in datacenters, and inference done on edge devices, often using specialized hardware. In the datacenters, a lot of NVIDIA's growth has come about because people worked out how to use GPUs for neural network training (and for Bitcoin mining). But there are other specialized devices that have been created, such as Google's TPU. Looking further out, Lip-Bu has an advantage over most of us. It might be an exaggeration, but not much, to say that no semiconductor startup is created without it crossing his desk, so just the "deal flow" gives a good perspective on where innovation is going. The five areas that he called out are: Silicon photonics Nanotubes Neuromorphic computing Blockchain Quantum computing To show that this is real, Lip-Bu introduced Rodrigo Liang, CEO of SambaNova. He talked about the eras of computing, but especially three trends: Memory technology (SRAM, HBM, DRAM) FPGA vs ASIC (which is a tradeoff between flexibility and optimality) New SW developments, especially ONNX (pronounced like "onyx") as a new neural network interface format for optimized models Lip-Bu introduced the CEO of another startup, Gopal Raghavan of Eta Compute. They are addressing ultra-low-power applications. He told us that "I started using Cadence in 1992, and I still do some layout for relaxation." His assumption is that the world is moving to a distributed computing model with edge intelligence, and that will require solving more than just power, but also security and privacy. But security and privacy are naturally more robust on the edge since a hacker has to break one unit at a time. Our brains split processing, with some signal processing out near the retina (vision) and cochlea (hearing). This leads to a remarkable dynamic range because this processing is extremely non-linear. This is very different from the current philosophy in neural nets, which is simply to pump the raw data into the network and make it bigger. Their approach is more like the brain, with a separate DSP and NN. Plus their low-power secret sauce that takes the voltage down to 0.2V where the Cortex-M3 consumes just 1uW (in 55nm). To wrap up, Lip-Bu pointed out that Cadence doesn't exist in a vacuum, and it takes an ecosystem to make the customer successful. Not just EDA tools and IP from Cadence, but also software, and standards. Tools from partners like Mathworks. Eventually, you need to manufacture your design, and except for a few companies that still have some fabs in-house, that means the foundries. Tom Beckley If you read last Friday's post, you'll know I titled it Open the Pod Bay Doors, HAL . Tom and I didn't confer at all, and obviously, he didn't put his keynote together since Friday, but in a weird alignment of the planets, Tom opened with the audio for exactly those few seconds of 2001. Tom looked forward to 2020, by which time everything is forecast to be a big number: billions of connected people, 25 million Apps, 25 billion embedded intelligent systems, 50 trillion gigabytes (I'm not sure what the SI prefix is for that) of data. This is the beginning of the 4th industrial revolution (4IR). This requires a lot more than just a single technology. Increasingly, these systems are "More than Moore" integrations, with multiple die from different processes in the same package. The heart of Tom's presentation was the new Virtuoso capabilities. I already covered this in a post that was published that morning Virtuoso 2018, a Fine Vintage , which looks at what it takes to design a modern solid-state Lidar system, which is one of the key enablers for autonomous driving, getting the cost under $100 instead of the current tens of thousands of dollars. Sign up for Sunday Brunch, the weekly Breakfast Bytes email![]()
↧