eSilicon’s CEO looks at industry consolidation, competition, China’s impact, and what’s changing in chip design.
Jack Harding, president and CEO of eSilicon, sat down with Semiconductor Engineering to talk about consolidation, business relationships, what it will take to survive in the IoT age, and how to better optimize chips. What follows are excerpts of that conversation.
SE: We’ve been looking at consolidation for a while and all these other macro trends. What are you finding is changing around you and within?
Harding: Several years ago, we began with the inevitable consolidation. There is more to come. But that is old news for the purposes of our discussion. I’ve got a slightly different angle, a by-product of the consolidations, coupled with an industry that’s not growing, nor is it forecasted to grow. The only way to be successful is to win share from other people. The industry always had tough folks in it, but I’m sensing business practices with an atmosphere that is a little less than a handshake atmosphere. There is less ‘co-opetition,’ a term I borrowed from Barry Nalebuff at Yale. There is more of an ‘I win, you lose’ sentiment in the industry. I don’t think that bodes well for the industry. Innovation comes from cooperation, collaboration, and a willingness to tolerate ambiguity about markets, growth, and opportunity. When people enter into business arrangements with a ‘I’ve got to win no matter what,’ and ‘You have to watch out for yourself,’ that will show up in less innovation and less invention, in the broadest sense.
SE: Is that a function of being in a transition period and a lack of food at the table?
Harding: That is precisely the problem. There is no growth, and if you are going to prosper and meet the expectation of your shareholders, which is the job of the executive staff, then you have to take it from someone else. We are in a society of ‘I’ve got to take it from you in order to win.’ It’s more like lifeboat ethics. The expression, “The rising tide lifts all boats,” is turning into, ‘You are not going to get into my boat.’ I see a big change.
Harding: There’s three dimensions here—technology markets, the entities themselves, and geographies. The technological curve seems to be intact for the time being, as the things we are enjoying now started five to seven years ago. And we haven’t seen the slowdown, which I believe is inevitable as R&D budgets are slashed in order to expand margins. We have to keep our eyes open for that. It will happen. With the dearth of venture/risk capital in the semiconductor industry, you aren’t going to get those large breakouts that we enjoyed for 30 to 40 years. Second, there is a whole other individual geographic force called China. Show me two investment bankers and I’ll show you claims of $100M of Chinese private equity funds ready to swoop in and buy your company. The Chinese are in the market to catch up and be less dependent on the Americans, in particular for semiconductor technology. Every board is hearing that at every meeting. It’s a dominant force in people’s thinking. The amount of time I spend being contacted by bankers inquiring if I want to be part of a Chinese rollup is growing quarterly. My friends at larger public companies also have similar incoming inquiries on a weekly basis about what they might want to sell or teams/consortiums they may want to join. It might be old news in some sense, in that a lot of people saw this coming, but on the other hand it is here and real now. Two years ago it was quite theoretical. Today it is an active force that can shift fortunes and strategies in our semiconductor space, both domestically and globally. Last, the marketplace—IoT as you made reference to it—is a giant question mark in my mind. I’ve talked at length about what that means and how it manifests itself, but I’m only starting to see the beginning indication that it’s a market that we can depend upon for growth and innovation.
The thing that gives me hope is that we’ve been at these transitional points before, and as an industry we’ve weathered through them and came out more powerful than we were going in. We can debate whether that has manifested itself in retaining earnings or not, but we continue to deliver miracles to the marketplace and amaze in what we can accomplish. The other thing is that the digitization of the consumer market continues to go at breakneck speed. Smartphones were the first leading-edge indicator of that phenomena, but we can expect to see that in all aspects of our lives. The prediction about automobiles being large consumers of electronics is real today and is accelerating with driverless cars and other trends we are talking about on a regular basis. Robotics, virtual reality, deep learning all consume acres and acres of silicon. There’s a robust growth opportunity there. The question is how do we prioritize these markets and our collectively shrinking R&D investments to access those things that give us the greatest return. But there is no limit to opportunity from a market perspective.
SE: Is this transition period really a repositioning of all pieces of the domestic and global economy? Business levels are connected, technology is connected, and everyone is playing in the same pond…except now it is really an ocean.
Harding: That’s a good observation and I agree with it. Twenty years ago Michael Porter from Harvard said there is no such thing as a low-tech industry, there are only low-tech companies. That is more true today than ever before. That was a prediction—today it is a reality. Every business is a technology business today. That drives the interconnect and mutual dependency upon the things we do as an industry, coupled with those things that were thought to be low tech or independent of semiconductor capability. I can’t think of an exciting new product that doesn’t depend on semiconductors going forward. I read an interesting interview recently between McKinsey and John Chambers. Mr. Chambers was making the point that the very evolution that you just described and the interdependencies for that innovation would cause a large percentage of the companies that are here today to be gone in 10 years. The implication was that their inability to evolve into this network of interdependent companies that rely upon deep technology in order to advance would hurt them.
SE: What has changed for eSilicon and how do you fit into this matrix?
Harding. We reconsider it all the time, although we’ve changed little in our 16 years. When we started as a fabless ASIC company, everyone thought that it was ridiculous, but three years later 80% of the industry was fabless ASIC, and for all the right reasons. When we went to market against our 25 competitors, we said we would build a chip from any combination of suppliers in the world. Everyone else had one recipe—one wafer supplier, one OSAT supplier, one IP supplier, and had to use their recipe. We said a working chip is not good enough—we want it optimized. We went down that path and, fast forward 16 years, there are now less than a half-dozen relevant players in our space. And probably only half of those can play at the bleeding edge of technology. We are happy to be among those three to four. The reason for our success is that we’ve stuck to that basic philosophy of optimizing the supply chain to give a superior result. It’s given us the opportunity to pick the best of breed at any point in time—who had the best solution for any part of that equation, that recipe. It’s challenged us in keeping track of a lot of suppliers and their offerings in order to be a credible supporter or recommender of their capability, but that challenge through automation has become trivial. Ten years ago it used to be more manual, whereas today it is much more automated. We find ourselves now where no one supplier can make everything necessary to make a complex ASIC. Every chip has third-party something on it. The value of the ability to select and integrate a third-party offering is a more valuable capability than owning everything that has inevitably, and by definition, at least one weak link in it.
Back to today’s original conversation, what trends help or hurt that thesis? We find ourselves reconsidering some things. We’ve invested heavily in some IP that we thought was mission critical going forward. Specifically, we have hundreds of people customizing SRAM. That’s paid great dividends to us. Five to six years ago, we were the first to get into HBM 2.5D technology, and today as a merchant supplier of ASICs, we are the leader in the world. We are working on projects deemed impossible three years ago. We are building chips of that ilk and scale today.
SE: So what can you accomplish now that you couldn’t several years ago?
Harding: We can walk into a customer who wants to make a 500mm², 14nm finFET-class chip, and break it in half and do half in 14 and half in 28. We can buy IP and put that little chiplet on the corner, put massive amounts of memory connected through an interposer and take that compute burden off. We have architectural options that were unheard of just two years ago. You can trade off notions of risk versus performance, risk versus power, risk versus die-size and therefore cost, manage time-to-market issues with existing technology that would be included in an HBM package. This is a whole other dimension of architectural tradeoff that wasn’t available before, even just a year ago.
Packing more on a monolithic chip geometrically raises more and more problems, while getting some benefits. You had to live with those problems. Now, there is another dimension in addition to the X and Y axis to make tradeoffs to approach a total solution. There aren’t two or three variables. The math underneath is solving 10 to 20 non-linear variables in order to come up with precise calculations on how a chip will perform around its PPA. Those choices, while it is almost impossible for us humans to get our head wrapped around it, are now possible today and we can rationalize massive amounts of data to come up with the true output of what those choices mean. When we decided HBM was where we wanted to invest our precious R&D dollars, we did a lot of work to predict the market when other folks were talking about other technologies. We went hard to solve individual technical problems, working test chip by test chip by test chip in order to solve signal integrity, DFT, and other problems. And today, we are sitting with a technology that continues our basic business thesis of providing technological choice.
SE: Are all these choices a deterrent for chipmakers?
Harding: Since we’ve automated the decision process to comprehend infinite numbers of variables into a complex ASIC, choice is our friend. It is the enemy of our competitors. It’s been an interesting educational process for our customers, as the vast majority of them have never considered that they had choices. Their supplier would tell them the wafer flavor they were getting. Their decision process was in very narrow constraints. For our customer today, we’ve demonstrated that we can give them infinite combinations of outcomes that we can rationalize technologically through automation and give them real-world examples in real-time. We can eliminate all the ridiculous cases immediately and narrow down to 10 cases, managing with tools and virtual tape-outs of those different combinations of flavors of suppliers. We can tell them standby power, die size, leakage, and everything about the chip. They can go to an infinite number of combinations down to those 10 variants that might make a choice.
It also puts us at the same side of the table with our customer. We don’t walk in selling one particular wafer from one particular supplier. Thus, we can say we are indifferent to the solution set that gets to the optimal chip. It allows us to give them unfettered, unbiased true advice about solving massive technological problems because we don’t have to sell them what we have on the shelf. That’s a huge advantage and a technical requirement now. If you are going to sell merchant chips that are customized like an ASIC, you have be able to say that buying a wafer from company X versus company Y has these pros and cons—and we are happy to do either—versus only having one wafer on our shelf to sell and it’s the only option to work with you.
SE: Are more customers expanding out now?
Harding: Yes, and it’s a practical outcome of customers spending $250M developing a chip. They should consider all the options. In the past, customers were focused on shrinking. The chip we are making today for people literally might have been 10 chips five years ago. They’d say these three we’ll keep the same and optimize these two. A complex ASIC might have 20 to 30 blocks on it. Before we start working with a customer, they start throwing 2 or 3 blocks of design over to us, requesting we do the physical design on these and then give feedback on these before we make decisions on the other 27 blocks. Those individual blocks might have been chips 10 years ago. Today, it’s almost pre-sales work to determine where the architecture is going and how to optimize. For example, when we tell customers that their 50 watt budget is going to be exceeded by 20 watts with the current way their chip is defined, we can go back and solve for that 50 watts and tell them how they could make it, with what supply chain, to meet that goal. The relationship between us and our customers has become much more intimate and collaborative, because we now have more information and methodologies that allow them to open their minds to a wide range of solutions that they would have never considered before.
More Consolidation Ahead
China and other big players will wait until the price of borrowed capital increases before making big moves.