Executive Viewpoint: Atoptech’s Jue-Hsien Chern

How does an engineer trained in building skyscrapers transition to semiconductor design? Surprisingly, quite easily.

popularity

What is the difference between skyscrapers and chips? Dr Chern has worked on both and he says it’s all about how you apply margins.

Jue-Hsien Chern started his technology career earning a M.S. and B.S. in Engineering from National Taiwan University and majored in structural engineering — bridges, dams, tunnels and high-rise buildings, all of which had to withstand earthquakes. That is a long way from being the CEO of an IC tools company in the United States, and we will start by telling the tale of his transformation.

chern

Chern came to the United States to finish his studies, still in structural engineering. With his passion for earthquakes, he either had to study in Japan or the United States. He did not speak Japanese, so “with my broken English I thought I could survive in the U.S.” He received his Ph.D. in Engineering from SUNY Buffalo and went to work designing power plants and heat exchangers. One day he got a call from Texas Instruments, which was looking for someone to apply numerical methods for simulations. Surprisingly they are quite similar in the two fields and this provided him with the transition from macro systems to micro systems.

Chern found this to quite an exciting transition. The terminology was different but the fundamental equations were te same. His time with TI exposed him to various aspects of the design process including SPICE simulations and analog design.

Semiconductor Engineering asked Chern about the similarities and differences between building skyscrapers and building chips.

SE: With very small process geometries don’t we have to overdesign in order to attain acceptable yield? Aren’t random variations the same as earthquakes?
Chern: When you design a building there is a strict building code that you have to meet and you have to get government approval. With a building, for example, you have an expected loading, and to be safe you multiply that by 1.7. Then you multiply by other margins.

With a chip you are making a tradeoff between yield and over-design. Management can make a decision. Do you want the highest performance even if you have to sacrifice yield or do you want yield and are you willing to sacrifice performance? How do you take into account the statistical distribution so that you can close timing but minimize the overdesign? This is why microelectronics is more exciting. You cannot blindly apply margin.

SE: After 10 years at TI, you moved even further into the micro space when you took a job at Technology Modeling Associates (TMA), which was focused on TCAD. You’ve said the mathematics at this level are still very close to those of building structures. (TMA was acquired by Avanti, which in turn was acquired by Synopsys.) How was this move?
Chern: When I made the transition I was naïve. I had no knowledge about anything beyond technical. Within TI, the only customers were colleagues and we would just talk about things. There was no commercial pressure. This actually provides a second way for an engineer to know he is doing good work. The first is through papers and the second is through customer usage. In a commercial company people actually pay to use it. This provides a certain level of satisfaction.

My role within TMA was VP of engineering. I later added the title of CTO. Initially I continued working on TCAD and added a few other products lines. Later I got involved in place and route. I talked to Wally Rhines, CEO of Mentor. He had been at TI as well, but I was a rookie engineer and Wally was a VP. I got to listen to some of his talks, but he never knew me. Eventually Wally persuaded me to come and join him and I went to work on analog, mixed-signal. This was an area in which Mentor was just getting started and I wanted to see if I could use my experience to learn how to grow a business within a larger company.

SE: So how did you end up at Atoptech?
Chern: Atoptech was a group of people I had known from the Avanti time. They had a product and their first few customers. I felt that it was a similar opportunity to TMA, but took it one step further and I became their CEO. I joined them at what was perhaps the worst economic downturn in our history. Everyone was retrenching. There were three things I needed to do:

  1. Continue to develop the technology and maintain our competitive edge. In a small company, once you have a breakthrough, and you gain attention, you want to retain the momentum.
  2. Find new customers who are willing to invest so that we can grow the company.
  3. Hire the people to grow the team. How do you convince people to join a small company during an economic downturn?

Thankfully, we got through the downturn relatively well.

SE: But Atoptech’s success depended on replacing entrenched tools. How did it go about doing that?
Chern: Our whole message was based on the shortest possible turn-around time. As one of our customers told us: Using this tool, our engineers can get to a good result quickly. This has one of two benefits — either the engineer can explore more design options and optimize performance, or they can achieve their design in a faster time. So our message was, ‘If you have product delays because you cannot close timing or fix the last few ECOs, then we may have a solution for you.’

One aspect to our success during the downturn was correctly selecting customers that fit our philosophy. We were focused on technology and support. Our objective was to make sure the customers finished in the right time with the best results. In order to do that, we had to look for customers who felt that P&R is part of their competitive advantage. Many people just see it as a commodity and any tool can do the job. They had to be willing to invest time and effort with us. We were fortunate enough to find those kinds of customers and they enabled us to mature, grow and develop new technologies. When I make a sale, I have to do it in a way that I can continue to invest. If I make a sale and cannot continue to provide technology that allows the customer to win, then in the end he will lose.

SE: Your product goes from netlist to GDS II. How do deal with companies that are providing physically aware synthesis flows?
Chern: For some designs, physical synthesis does provide better results, but not for all. You can make the argument that synthesis and placement tied together is better but by a similar argument you may find that placement and CTS together is better. You could argue that you should be doing synthesis all the way through routing, but this may become too complex. Historically, EDA puts in a level of abstraction to constrain and segment the problem into a more manageable problem and provide optimization at each stage.

For a design, be it physical aware or not, what we can say is that if they have a better netlist, then we will get a better result.

SE: What new technical challenges are you facing today?
Chern: We see three trends, all driven by the migration of the technology node. We all speculate about when Moore’s law is going to slow down, but judging by the past three years, it is not slowing down. It may even be accelerating.

Not long again there was a big hype about DFM. Today, this is just a part of the design rules that we have to work with in the router. This was a discontinuity in the technology, which required a rewrite of routers. This presents an opportunity for a new company.

The second area that is becoming more important is statistical variation. With the explosion of many corners you have to look at all these corners and yet still not over-design. Over-design will kill all of the margins. The challenge is dealing with these statistical variations inside of the optimizations of your P&R tool so that you can close timing with the best possible performance, power and area.

The third area is the complexity of design and this will only get worse. In true SoC design, there are two areas of impact. The first is at the block level. In the past, a design was built using one design style. Today, blocks come from many companies and are in many different styles. The second area is that timing closure at the top level is becoming more challenging when you stitch together many of these design blocks. The traditional ways of timing budgeting becomes very cumbersome and inaccurate.

SE: Recently, power has risen from being a secondary consideration to being a major parameter for optimization. How does this affect P&R?
Chern: It has a major impact. We need to be able to handle very complex power domain issues and support for things such as CPF, UPF to ensure the design is correctly represented in the physical implementation. The second area is that it creates very complex clocking schemes. Both of these add to the problem complexity. Plus you have things like variable voltage and frequency.

As a tool vendor, we have to try and take away things that an engineer has to be aware of. The key to design automation is taking things away from the designer that they used to have to do by hand. The user experience is important. We have to ensure that the algorithms work in a way such that the engineer does not feel like he is fighting the tool. They can focus on fighting their designs.

SE: How large is Atoptech and what is your strategy?
Chern: We grew to being over 100 people some time ago and have been profitable ever since I joined the company almost four years ago. There are three questions that I ask. The first question is, ‘Do we have room for growth?’ I believe that we do because P&R is one of the largest growing segments in EDA and we only have a tiny market share today, so we have plenty of room for growth. The second question is, ‘Do we have the ability to grow? Do we have the technology and people to grow?’ We have shown over the past few years that we are one of the few companies able to keep up with the change in the technology from 28 to 20 to 16 and 14 with FinFETs. We are already certified by TSMC for FinFETs. Thirdly, ‘Do we have the resources for growth and we have been able to grow from our retained earnings?’ This means we can build a sustainable, growing company for the long haul. Until we have to answer ‘no’ to one of those questions, we don’t have to change our direction or focus.

SE: What advice do you have for a young entrepreneur?
Chern: My advice for people looking to start a company is first, don’t over-plan, and second, seize the opportunity. It sometimes takes a leap of faith, and that is what you have to take. You have to find the new opportunities and new challenges. By over-planning, I am talking about not worrying too much about what will happen in three or five years. In EDA this is an eternity.

[Editor’s Note: In June 2013, Synopsys filed suit against Atoptech related to a license agreement that Atoptech had with ExtremeDA, a company that Synopsys purchased in 2011 and the same year in which ExtremeDA licensed technology relating to a method for parametric on chip variation POCV to Atoptech. That agreement came with certain conditions that Synopsys claims were broken.]