A candid interview with Atrenta’s CEO about industry consolidation, growth opportunities, and some of the challenges facing midsize EDA companies.
SE: What keeps you awake at night?
Bose: What I worry about more than anything else is the need for us (at Atrenta) to show growth on an ongoing basis. A company’s challenges change with the lifecycle of that company. In the early days you worry about survival and trying to establish yourself in the industry. Fortunately, Atrenta is a bigger company today, so the nature of the concerns has changed. Today what is expected of us is to show strong growth, on a consistent basis, and to shine within the industry as a fast-growing company. Unfortunately, this is an industry that is not growing very rapidly, so that’s not an easy task.
SE: The problems that people have to solve are bigger, too – and more complex. We’re dealing with shrinking features, integrating IP, pushing into 2.5D and 3D. Are you positioned to handle that?
Bose: There are a number of forces driving challenges for the industry. They generate more work and they create more opportunity, but they don’t naturally come with a nice financial solution. These are two different things. You develop products, you solve problems, you have to take them to market, establish value in them and monetize them in the end.
SE: And you’re taking chances that it will materialize, right?
Bose: Yes, exactly. The two don’t always connect in an easy way. In terms of the challenges driving the industry today, the smaller technology nodes generate a lot of problems. There are issues with power, test coverage has to be higher, timing closure is a bigger problem, congestion is a problem, and so on. That’s one force that is driving innovation. The other challenge is growing complexity. Although there is a relationship between smaller nodes and complexity, complexity brings its own problems. Initially, it makes the verification problem more significant because this problem grows in a non-linear way. For example, if you double the size of a chip, the verification problem may go up by 4X. This verification complexity effect creates new challenges, and the techniques to address the problem don’t scale, so you have to reinvent new techniques to solve the problems. Additionally there are business challenges. For example, you can design the chip, but can you afford the cost of developing it? Also, can you monetize and recover these costs, and does it make good business sense? There is a lot of room for reducing cost; including improving productivity to reduce the amount of time it takes to design a chip. That’s where IP signoff and SoC efficiency play into these issues in a big way. These are some of the multiple challenges I see for the industry.
SE: What about stacked die and its impact on power?
Bose: Well, 2.5D, 3D, and some of the power related issues are slightly orthogonal. Some of them are driven by the fact that at smaller nodes the cost of the design and specifically, the cost per transistor, is going up. This is the first time in my life that this has happened. People are struggling with what to do about it. Stacked die is interesting, but it is not yet financially viable. But the solutions to that problem involve improving manufacturing techniques, getting better yields, implementing through-silicon vias differently to reduce the cost. There are a number of techniques and these are all attempts at handling complexity at an affordable cost. That solution has to come from people who are on the manufacturing side. But if you talk about leakage and dynamic power, there are design techniques to improve those areas.
SE: Where does Atrenta fit into all of this complexity going forward?
Bose: Atrenta is one of the larger private companies in EDA today. We have, in some ways, tried to redefine the problem in a way that we think makes sense. We look at issues up front in the design cycle, addressing problems at the RTL. We have developed a platform in which we look at multiple domains of problems, such as power, testability, clocks, constraints, and physical. In the beginning it was a missionary approach because it was a new concept in the industry. But as new technology nodes, complexity, and IP have proliferated, what we do has become more relevant, more significant, and more of a pain killer. As a result, we are in a good position. The new concepts we defined and set out to accomplish early on, are becoming more relevant and contributing to our growth. We’re now used by over 250 companies and we are an essential part of an IP or SoC design.
SE: Where do you hang your hat as a company? Is it a company that provides tools for integration? For the leading edge? Or do you push into new areas such as security or global problems such as power?
Bose: Atrenta has taken a platform approach. We try to make a very comprehensive solution starting at the earliest stages of design. This does span multiple areas, with power being one of them. We also have a layered approach. There is a lot of what we refer to as foundation technologies. These are a set of analytical tools that work on the design and extract information about the design and help identify problems and areas of improvement. You can take that platform and apply it in different areas. You can look at the result of an analysis and see what it tells you about the testability of the design. Alternately, you can look at the result and see what it says about power consumption. While you’re at it, you can look at how to improve the power consumption. The platform approach gets targeted toward different problems. As you can see, we hang our hat around the platform. That’s what sets us apart and makes us unique. We’ve defined and implemented this concept and now we are finding more uses for it and solving more problems.
SE: The customer base is changing, too, right? Now companies are more open to new solutions because the problems are so complex. On top of that, there are new companies coming into the market. They don’t have any tools or internal legacy tools to deal with.
Bose: The internally developed tools have almost gone away. I used to be a part of that in my Bell Labs days. Except for very few companies with internal development, most people rely on commercial tools. It’s encouraging that systems companies and what we have historically thought of as software companies are getting into the chip business. They are focused on large, complex SoCs. What they’re trying to do is integrate multiple functions and applications. They use it as a way to leverage whatever they’re offering. The iPhone, for example, has so many functions put together by IP that are integrated into the device. But these companies acquire somebody or some technology. It’s not as if they’re starting from scratch. They have expertise, but they are all fabless and nimble—they can move from one generation to the next very quickly. I see that as a very interesting trend. There are newer companies with the Internet of Things and wearables. This is all good for semiconductors, and it’s good for EDA. What’s most interesting is that companies that were not historically chip design companies are seeing enough value in the SoC. It’s a way to bring leverage to whatever it is that they’re offering their customers.
SE: Do these fabless OEMs have an advantage over companies fighting for a socket because they know what they need and they can design it?
Bose: Yes, and that’s one big benefit. However, there’s another big benefit, too. If you look at how the value gets distributed through the chain, it’s heavily concentrated toward the end product. Then it starts dropping off as you get lower in the chain, such as into semiconductors. The companies that are driving the trend and the ones making the decisions have more flexibility and are less constrained with having to win a socket. They are driven by creating the best possible product with the flexibility and the wherewithal to do it. This is a good benefit.
SE: What effect will systems companies getting into chip making have on the industry?
Bose: It’s going to put a lot of pressure on the semis, but it may not be a bad thing for EDA. These are a new breed of customers, and they’re less constrained by history and by some of the challenges that face the traditional semiconductor companies. There’s a clearer line of sight to doing something in their design flow to improve their business.
SE: What’s happening in the Chinese market, where traditionally they have not used many EDA tools?
Bose: China, from Atrenta’s perspective, has four segments. There is a huge amount of multinational activity, which is a large area of our focus. Most semiconductor companies have large design teams in China. The second group consists of the government labs. The government labs are a catalyst in promoting semiconductors in China and they do buy tools. They can be a sizable customer. The third group is comprised of the larger Chinese companies, and they buy tools like any other big chip company. They are growing and maturing but don’t yet have the large company approach to design and EDA spend. The fourth group encompasses the startups, and China has hundreds of them. They don’t buy too many tools from us since they have the support of government labs and other sources. In terms of our mindshare, multinationals are the largest, followed by government labs, the larger companies and then the startups.
SE: We’ve seen a lot of consolidation in EDA these days. Is it difficult for a midsize EDA company to remain independent?
Bose: I wish there were more midsize companies. That would make the industry more dynamic because you could go to multiple places to get solutions rather than just a few. Midsize companies are largely disappearing. For Atrenta, the possibility of doing an IPO is quite real and this is something that we consider as a serious option.
SE: We haven’t seen an IPO since Magma.
Bose: Yes, it’s been about 12 years. Our thinking is that to do an IPO, you have to do it from a position of strength. Just creeping into becoming a public company isn’t good enough. In a good economy, in good times, in a good market, that might work. But it is not something that can withstand a bad market. That’s why most companies don’t want to test the waters to see if they can be the next large EDA company. For us, we have to get to a position of strength and we’re on track to do that. We’re growing quite rapidly. If we sustain this growth, and there are many indications we will, that does become a real option. But I do wish there were a few more. If you think back to Avanti!, Verisity and Simplex, that was a nice collection of companies that offered solutions and choice. And it wasn’t just for tools, it offered options for employees as well.
SE: Can big companies compete with smaller companies on products?
Bose: We talked about the new challenges that are coming up. Historically, the solutions for these challenges have come from startups. Startups build a solution, get to market, and they do whatever it takes to deploy their tools and grow. Unfortunately, the number of startups is shrinking. What that means is that solutions need to come from the larger companies. The larger companies are paying attention to these problems, and there is activity in these companies to come up with solutions. However, coming up with a product is one thing, while taking it to market is a whole different game. In fact, coming up with a product is the simpler challenge. To succeed in the market you need the focus, the hunger, and a realization that if you don’t sell it you’re not going to survive. In a startup, you don’t have other products that will carry the company if this product doesn’t sell. These are very powerful forces that lead to solutions emerging and coming to light. It will be interesting to see how the solutions emerge out of the larger companies.
SE: But can big companies justify investments in some areas?
Bose: There is some element of that. Unless the market is seen to be big enough, or have big enough potential, it is unlikely to get their attention and the resources. More importantly, I think it’s the battleground of taking the technology through the early adoption phase. Nothing ever comes out in this industry and works the first time. It can be a long period, or if you’re smart, it can be shorter. However, there’s always a period of sheer struggle to get a new product out and into maturity. How do large companies go through that phase? It’s not just someone in management or R&D saying they’re going to develop a product. It’s whether that product will get the support of sales and the rest of the chain to make it to maturity.
SE: We’ve been rewriting some tools for more than 20 years. Will there be a point where they won’t work anymore—where they can’t be tweaked through multi-core or multithreading or better algorithms—and something completely different has to come along?
Bose: The classical approach to this problem has been to use either some new level of hierarchy to let you deal with complexity in a manageable way or, more recently, it’s been a driver for using IP as building blocks. Blocks are a great way to partition a problem. We certainly promote the notion that design should be done in blocks due to many benefits. First of all, it gives you a unit that you can focus on, get your arms around, and make sure it’s tested and proven to be right at the block level. Once a problem is found and fixed, it should not surface again. Then you need some intelligence in the way you deal with interaction between blocks. This reduces the problem of a billion gates down to a 1,000 blocks that are interacting. This whole notion of block-based design does lend itself to this approach. There is no fundamental limit to why this approach would ever stop working. There is a limit that you can only place and route so many instances, and this will limit the block size. However, the industry seems to be able to cope with that. I think the notion of dealing with complexity with some form of partitioning, is a fairly elegant, well tested and proven idea which has its roots in the circuit board days with pre-existing blocks that were reliable and proven. That model is proven to work. Business problems are a big challenge. How many chips can you afford to design at 20nm or 16 nm? There are other frontiers that are getting more complicated, such as software becoming a growing part of the SoC. Data shows that the cost of software in an SoC will soon overtake the cost of hardware in an SoC. Additionally, verifying hardware and software together is another issue. That’s why you’re seeing a boom in emulators. Everyone is trying to verify software with hardware.
SE: Where do you see the idea of design platforms?
Bose: About five years ago I thought the notion of platforms for a chip would take off more rapidly than it did. Design platforms are such an elegant concept, but their progress has been slower than expected.
Twenty years ago I thought block-based design was going to happen imminently. It took a long, long time and even today, there are still some groups that don’t buy into IP blocks and reuse.