Lip-Bu's Fireside Chat with Ed Sperling—With Real Fire

Usually the phrase "fireside chat" is just a figure of speech, but Wednesday's came with a real fire. A spotlight caught fire above Mentor's booth, and the entire conference center was evacuated for half an hour or so. Luckily, we got the all-clear and Lip-Bu's chat with Ed Sperling started only 15 minutes behind schedule.

Ed opened with a gentle question about what is unique about automotive electronics. Lip-Bu said reliability and safety. He drives a Tesla with Autopilot, as close as we have to a commercial self-driving car today, and it is "great fun". The big difference for the semiconductor ecosystem is that we haven't really had to deal with those requirements before, and the car companies are struggling with the same issues since they have relied on the Tier-1s before but have come to the conclusion that they need to design their own chips. Of course there are a lot of software requirements. Security is very important, too. Not to mention that the chips have to last 15 or 20 years, not like the two or three years that suffice for a cellphone chip.

Ed wondered if the customers need to do more with the tools or do we need more powerful tools. Lip-Bu said it was a combination. It is new for everyone. There are lots of requirements including the ubiquitous ISO 26262. Some functional safety verification needs to be embedded in the tools.

What about "We are going to be sued"?

Lip-Bu laughed. "Sure everyone is trying to pass the risk to someone else in the food chain. All we can do is proactively address the checklist and share the responsibility. The auto vendors, the actual car manufacturers, are feeling their way since they have not had to deal with us before, in the past they operated through the Tier-1s. He had had dinner with a friend who is head of a car company and found that after a few hours of discussing there were already lots of things that the two companies could work on together.

Ed switched gears to the consumer space and, in particular, issues with power.

On the tool side, Lip-Bu said, we are improving how we switch things off with approaches like power gating. But with his Walden hat on (Lip-Bu is not just the CEO of Cadence but is the founder and chairman of Walden International), he has just invested in a company doing near-threshold-voltage (below 0.3V) asynchronous design. The idea is to design chips where the battery lasts 5-10 years. Another potential investment he is talking to is doing energy scavenging, with no battery at all. So the Cadence tool side is one thing, but also new material and techniques come into play.

"What about medical?" Ed asked. It always has promise and never seems to arrive.

Lip-Bu disagreed. Things are already happening. One of his companies does microfluidic optical development and use big data approaches to get analysis done in six hours. He told them not to outsource their data, but build their own software team. The data is very valuable. There is lots of scope for machine learning. The whole medical field is important and the margins are better than in IT. It is starting to come together now. Via technology does 3D image sensors using a cellphone. No need to do blood tests for insulin with a needle, just wear 3D sensor. These are things here, not in the future. It is all just beginning.

Ed was frustrated that he has so many remote controls and his devices don't communicate with each other.

Lip-Bu sees the home as another exciting big platform. On the Cadence side, we work with customers driving unique requirements, and privacy in particular is very important. Home security and robotics with AI built in are big potential growth areas. Lip-Bu has two examples he has invested in. Canary can do home security and you can be anywhere in the world to interface to it. In the robotics space, Rokit which has lots of intelligence, can recognize family members and automatically learns. It alerts automatically if someone unauthorized shows up. There are also remote controls with beam forming to control large numbers of devices. That's another product that is already out, not in the future.

The cloud is changing with so much data coming in so need changes for better performance in storage and networking. There is a lot of ASIC and SoC design for hyperscale web service companies who are investing a great deal of money. As Lip-Bu put it, "We are working hard to please our teenagers. They want instant response and we have to work hard to provide it."

"So we are rethinking how computing is done," Ed said. "How does this affect design?"

On the tool side, Lip-Bu replied, we have a couple of areas we are excited about. Time to verify is becoming critical. Run time is a big part of that, every 2X, 5X, 10X is huge. The verification depends on the type of design (automotive, home, cloud, etc). Simulation, emulation, formal, what Jim Hogan calls the three-legged stool. But it is not just digital, we need analog and digital to be verified together. Of course, we have customers pushing to 5nm, 7nm but there's a lot of design in other nodes.

"We know who the customers moving to 5nm and 7nm are," Ed said, "they are household names. But what about old nodes. How do you serve that market?"

Lip-Bu said that a big chunk of our business is analog/mixed-signal in things like industrial. They are comfortable at 45nm, 65nm, and even 350nm. Some of what we learn at advanced node we can provide back and work with them to improve their productivity and performance. Then they buy more of our tools and IP, so it is a virtuous circle. They also look to us to help them decide whether to stay at 28nm or move to 7nm and decisions like that. Many customers treat us as a partner, not just an EDA vendor. They ask “You know all this stuff, what are the tradeoffs?” Along the way we can help them because we can tell them what their customer's customer needs. That differentitates us.

We work as a team. I end up talking with the CEOs and heads of engineering. But we have other team members to talk to the CAD teams, the engineerign teams. We have a very holistic way of working with the customer.

IP is a very interesting business, too. Dealing with SoC, there are a lot of IP blocks and big companies used to own their IP. But protocols keep changing and so it is harder and harder. So a bit like EDA 25 years ago, it is moving from internal to third party. Design groups can deal with bugs in EDA tools, but not with bugs in IP since they are designed into the products. We are still learning how to be a really good silicon proven supplier and drive the standards. Double digit growth for us. Growing organically and acquiring. Started with memory with Denali, Tensilica for machine learning ADAS, and so on. Some IP we are building and some we are acquiring. We listen to customers and they tell us which to buy, which not to buy.

Ed's last question was that hardware companies are being told they need to develop much more software to win a socket. There are more software engineers than hardware engineers on SoC projects. He wanted to know how that changes things.

Software is everywhere. Software defined networks. Software defined everything. Almost all Cadence's customers hire more software engineers than hardware. One big problem is that the best software talent doesn't always come into the semiconductor industry. "Even my own sons, with engineering degrees, picked software." Another big problem is that the ROI for software is not good. It still helps get the deal at the semiconductor company but the customer expects the software for free. However, it is also an opportunity. Customers like NVIDIA and Qualcomm increasingly define themselves as system companies and are have more success at monetizing the software effort that they are putting in. We still have to learn as an industry what works and doesn't work.

If you missed the Denali party, or couldn't make it to DAC, here is a taste from Tuesday night. YMCA with Disco Inferno, and the crowd doing all the actions. If you want to know how it started, read Party Like It's 1999—How the Denali Party Started.

Next: Securing the IoT for Billions of Possible Intrusion Points

Previous: DAC News, Wednesday