Get email delivery of the Cadence blog featured here
Monday was Lip-Bu's turn for a chat with Ed Sperling. The same event last year was billed as a "fireside chat" but was postponed for 15 minutes when a spotlight really did catch fire. Taking no chances, this year it was re-branded as a one-on-one chat, and the spotlights obliged by not bursting into flames.
Ed started with a softball question about which markets are going to be hot. Lip-Bu gave the stock answer (mobile, automotive, IoT, HPC), but emphasized that there isn't something on the horizon on the scale of mobile with literally billions of units per year shipping. But he sees lots of opportunity and disruption in cloud datacenter infrastructure due to big data. Adding data from cars to hyperscale web services and it's huge. Lip-Bu is excited about the whole semiconductor industry.
As to what that means for EDA, a lot of markets are much smaller but chip volumes don't directly affect EDA. Ever since Meera pointed out that Lip-Bu says "machinelearningdeeplearning" as if it is a single word, I can't help noticing it, too. Lip-Bu pointed out that machinelearningdeeplearning will be folded into our own tools to make them faster and more efficient. There is a huge change in customer requirements as we move up from traditional EDA to System Design Enablement (SDE). Part of that is to optimize systems for advanced 2.5D and 3D packaging: system analysis, modeling, and simulation. It's not possible to wait until tapeout and the discover the power is too high. Advanced nodes really do have to be first-time correct, especially since the manufacturing cycle time is so long. IP is becoming more and more important since it is too risky to use building blocks that haven't been silicon proven. It is now the case that 40% of Cadence's customers are system companies and service providers optimizing the solutions that they need. This is all taking a lot of innovation, as evidenced by 23 new products developed organically inside Cadence in the last three years.
Ed picked up on 2.5D and 3D advanced packaging. "You guys were 15 years ahead of the market, is this becoming real?" A few years ago Lip-Bu talked to several companies all saying 2.5D and 3D is great, but who will be paying for it. We worked with TSMC and tried to get ready for customers, but the customers were not yet ready. "This year, during the Christmas break, I was skiing in Utah and a leading packaging company wanted to meet me. Customers are now chasing them. There is huge demand from the customer side. For some applications like high-speed SerDes, and silicon photonics packaging is critical."
Ed is worried whether the tools and methodologies can handle it all? Lip-Bu said that we are not just pushing to advanced nodes, although obviously we are doing that. We have active programs with key leading professors like Andrew and Jason [presumably Andrew Kahng at UCSD and Jason Cong at UCLA], and others in Israel and China.
What about startups? Cadence has a capable team with machine learning to really drive PPA involvement. "But we are also engaged with professors to see if there is new technology we can bring in. But I am also a VC and not many people are investing in semi, so I get to see 1500 business plans a year. I read each one and get insight into what are the new things coming. Along the way we keep and eye out for opportunities like Rocketick, which is now integrated into Xcelium and the customers love it."
Of those 1500 proposals, how many are good? "I pay attention to about 30-35 that we really deep dive into, and most VCs don’t invest in semi or IP at all. Walden International invests in about 25 companies per year."
Ed is worried if we have the manpower since the hardware industry been losing engineers, but software has been exploding. "A lot of professors in some unis think semi is a sunset industry and I try and convince them it is not. Everything is software defined from network switches to automotive, so software becomes more and more important. Tensilica is perfect for machinelearningdeeplearning, for automotive, for on-chip genomic sequencing. But I have two boys in engineering, and they both decided to do software. I try to convince them EE plus CS is the holy grail."
If you look at Facebook and Google, they are doing their own hardware, too? Are they now competition for EDA? "No," said Lip-Bu. "They are my customers."
How will machine learning and AI play out? "Machinelearingdeeplearing is a high priority for me. Hyperscale web services, ADAS, genomic sequencing, EDA. A new class of hardware and software will drive machinelearningdeeplearing and the applications are very broad. I keep track of 35 startups and a few I invest in. But Google, Amazon, Microsoft each have 1000 people working on this and will need a lot of EDA tools for this. And we also are using machinelearningdeeplearning within our own tools across the flows to make them faster and more accurate. We are seeing some early success and we are making progress. Customers love it. I’m excited. Today, all our R&D is ported into the cloud, so we can use massive stuff on Amazon. Our CIO is really good to scale things like Pegasus up to 1000 CPUs."
Ed wanted to know if autonomous driving will roll out the way people think and on the timeframe people thing. "We are at level 2 for most advanced cars, so a long way to go. I drive my Tesla from Piedmont every day and get through 50 emails. I can’t wait for level 5. Currently autonomous driving is level 2, your hands still have to be around, I’ve been investing in level 4/5 at Walden and just made an investment. It's hard to do. The trend of moving to level 4/5 is faster than we thought. Back to looking at EDA for automotive, there are some requirements such as functional safety and also some sensors. Tensilica plays a key role since it is programmable and low power. It is a great opportunity, but it takes time. It is not going to be like mobile turning on overnight. Some other verticals like datacenter have big capex. Industrial IoT is real with companies like GE and Siemens. EDA needs to transform itself into System Design Enablement. In the end it is application driven."
Every time I see an interview with Ed, he asks about medical, because it seems to be always on the brink of big things, but has been for a decade. "Where are we now?" he asked. Lip-Bu thinks it is an exciting space. "Cancer is not one drug cure-all, it is patient specific. That means genomic sequencing and using simulation from EDA to accelerate genomic sequence analysis. We are just at the start of patient-specific data. The analytics have tremendous value. EDA is not stopping at the silicon, it applies to automotive, to medical."
If you have ever seen a Lip-Bu keynote, you know that he likes to talk about the things he has seen that he is excited about. "In some ways this is the most exciting time ever for the EDA business. I haven’t sold a single share. We can do a lot in automotive, IoT, medical, defense, the whole datacenter. This year, semiconductor is going great. A friend of mine managing a semi fund has 38% return this year. NVIDIA is almost $100B company, Broadcom over $100B market cap. EDA is transforming into SDE for the system and service providers."