Nitin Dahad talks to Graham Curren, CEO of Sondrel, who founded the chip design service company nearly 20 years ago. We talked about why he started, their first chip design, what’s changed over the last 20 years, and about what’s in store for chip design in the future. Here’s one clue: it’s all about the size.

[FULL TRANSCRIPT AVAILABLE BELOW]

Welcome to Embedded Edge with Nitin, a podcast show hosted by Nitin Dahad that brings to life to the stories behind the embedded systems, technologies and products. It’s the show where you’ll hear from both engineers and executives on some of the topical news and challenges in the world of embedded systems design.

Here’s your host, Editor-in-chief of embedded.com, Nitin Dahad.

NITIN DAHAD

Welcome to this special edition of embedded edge with Nitin. I caught up with Graham Curren,  CEO of Sondrel, a provider of complex system-on-chip design, offering a turnkey ASIC service from designing the system to supplying the silicon.

Graham founded the company some 20 years ago. I asked him about the background to why he started the company, some of the early challenges and customers, and what have been the key trends and changes in embedded systems design over the last 20 years. The company has also worked on many very large chips, so we asked about some of the challenges, as well as what he sees as the next key challenges in the semiconductor industry.

NITIN DAHAD

Hello, I’m here with Graham Curren, CEO of Sondrel. Hi Graham.

GRAHAM CURREN

Hello Nitin, good to talk to you again. Nice to see someone after so long.

NITIN DAHAD

Exactly, so I thought we’d cover a little bit about what you’re doing with Sondrel, but before we do that, let’s start with a little bit of background. Sondrel’s been going since 2002. You founded it then. Tell us a little bit about the background and why you started. Obviously, things have changed in the last 20 years,

GRAHAM CURREN

Yes, it’s 20 years next summer, so it’s been a quite an interesting journey. Before Sondrel, I was working for Avanti, an EDA company, and we were largely specialized in physical design, extraction and so on. And at the time I was feeling that it was becoming increasingly difficult for small companies to do their own chip design, and particularly around the physical design areas because it was becoming very expensive and very specialized, and what I was seeing, particularly among startups, is that they would decide they were going to do a chip, take a year to build a team, take a year to work out what it was doing and how to get the flow and everything working through the chip. And then spend another couple of years with the team sitting around waiting for some sales to come through.

So, spending a huge amount of money and it just seemed to me that if you compared those types of work with the economies of scale that we were getting at the bigger semiconductor companies, there was an opportunity for somebody to come in and say, well, we’ll do your chip for you, we’ve already got the methodologies and the processes, and we know what we’re doing.

Unfortunately that somewhat coincided with the dot com crash and the drain of venture capital out of semiconductors never quite went in the way I expected, but nevertheless we took, from day one an international approach and working flexibly across multiple markets in multiple areas and therefore over time we’ve managed to grow substantially. We’ve had a year-on-year growth averaging sort of 20%. A couple of downturns but mostly I think we’ve had two years when we downturned, but every other it’s gone up, so it’s gone pretty well.

NITIN DAHAD

And what remember of your first design, your first customer? Is it something you remember, or you can tell, or is it so far in the distant past now?

GRAHAM CURREN

So, it’s far enough away that I think the confidentiality restrictions have gone, so the first customer was Infineon in Germany. There were three of us originally and we spent – at least two of us – time flying backwards and forwards to Munich and spent six months working during the week in Munich on site. The thing I remember most about that was it was a very cold winter, and it was like minus 16 degrees centigrade every night. So, it exposed me to very cold weather for the first time, but it was a lot of fun actually. I’d been a manager at Avaniti before that for quite a long time, and I hadn’t really touched the tools, so getting back to actually touching the tools and finding out what was real was a great experience and I really enjoyed that.

NITIN DAHAD

And what are the changes you’ve seen over the last 20 years? Tell us about the top changes and trends from then to today.

GRAHAM CURREN

Well, I probably go back even further than that, but you know, since timing driven design came in, synthesis came in, I don’t think there have been any major changes, or anything really drastic. There’s been some smaller things that come in of course – changes every now and then and an increment.

But the problem I see today and the challenges I see today are not that dissimilar from the ones that we saw 25, almost 30 years ago. I’m old enough to have been involved in early Cadence tools, when Cadence first came onto the market and so. There’s been a number of people in my generation that have been through that. Challenge are very similar, and particularly when Avanti brought in timing driven designs and Synopsys board in synthesis. There hasn’t been any major revolution since then that I’ve seen.

NITIN DAHAD

That’s interesting. Maybe that leads us onto the next thing. You sort of evolved from design services to developing ASICS and IP platforms. Tell us a little bit more about Sondrel’s business model, and how you operate.

GRAHAM CURREN

So right from day one I had – like many people starting their own company – a fantastic hockey stick business plan that said “we’re going to be at a billion dollars within two years” or something like that. Life turns out to be somewhat harder than that, but there has always been an intention to do silicon supply as well as design. For a number of different reasons that decision has taken some time to come to fruition, but we decided a few years ago now that that was the right time.

We needed to build a good capability to actually deliver chips because these are big, complicated chips that we work on. It’s not a small device that can be handled easily. They’re expensive and the testing is very difficult and complicated. You need to have a lot of capabilities, so we needed to have the scale and the infrastructure to be able to handle that.

So, we launched our ASIC business a few years ago, and it has shown a really good uptake so far. It’s got a lot of the advantages that design services doesn’t have. It’s much more longer term. It allows you long-term engagements, which is really the big attraction for it.

NITIN DAHAD

And I think I think you’ve launched a few IP platforms as a result of that.

GRAHAM CURREN

Yes, so one of the key things here is to how you get your design work that you do on one design to allow you to be more efficient on the next, whilst of course maintaining customer confidentiality and differentiation between the different products. And companies have always tried this in all sorts of different ways, whether it’s having IP blocks or soft IP or hard IP. Whether it’s platforms or subsystems or whatever it is. So we’ve come up with this approach that we call “architecting the future”, which allows us to address four or five different markets, and we’ve really focused our engineering efforts on how we assemble those complex SoC’s efficiently.

We don’t for example design our own hard IP blocks and then tell our customers that’s what you’ve got to use. What we do instead is work out how to really efficiently integrate other people’s IP into the systems and make sure that they’re well characterized and well known and that they can be done at low risk, and quickly.

NITIN DAHAD

And you’ve got the manpower behind that to put all that together? It still requires manpower, even though we’ve had the EDA tools people talking about putting AI and intelligence into designing that, but you still need that manpower, I guess.

GRAHAM CURREN

Yes, it takes a lot of background effort to design a chip today, you have to work out what tools you’re going to use and what order you’re going to use them. You have to train people. These will take a lot of time, but then also you need to do all this R&D to develop the platforms or whatever else you’re going to do. So, like all companies in our situation, we have a number of people who have to be in the background if you like not working on customer projects. And this is a really, really important part of our business. This is what drives our future revenue and our future growth, which is our ability to develop our own IP. Something that will convince the customers and enable their projects to be successful. So we spend a lot of our revenue, many millions of dollars every year, on internal R&D.

NITIN DAHAD

So, what do customers typically come to you for, and what’s the type of customer that comes to you? Is it a for a particular market or application or is everything?

GRAHAM CURREN

It can be anything actually. We always look at this and think, should we focus on a particular market or something, but our expertise is in delivering complex designs in new technologies. So we don’t typically engage in designs which are 40 nanometer or so on. Nearly all the designs that we work on are 12, 7, 6, 5 nanometer; we even have one on 3 nanometers. So that’s our sweet spot and that tends to indicate which markets you go for.

So, the markets we address tend to be in networks, imaging, AI, those types of markets or application areas. It in terms of companies, then it can be anything from major semiconductor companies all the way through to startups.

NITIN DAHAD

I think I did say earlier, but large chips seem to be your specialty, and that serves that market, right?

GRAHAM CURREN

It’s not really. Although we do talk about large chips, and of course, one of the reasons we talk about large chips is because they’re more interesting for people. You don’t tend to do a press release saying I’ve just completed a small chip, I think nobody wants to read it. The reality is, yes, we do some very large chips. We’ve taped out a number of chips which have been well over 600 square millimeters, which is the reticle limit in 7 nanometer; we have one in in 5 nanometers that’s that size. These are many billions of transistors.

But we also do some more normal size 60-70 square millimeters and even in some cases 2, 3, 4, 5 square millimeters. But the common factor here is that they all have something that is unusual about them. I mean, there are really only about four companies in the world that can design in these technologies.

So that gives us a strong differentiation. And of course, it gets our engineers interested. They love doing the stuff that’s in the new technologies and I’m an engineer by background, as I’ve already said, so it’s much more interesting to do complicated hierarchical heterogeneous design in five nanometer than it is a small mixed signal design, the same sort that you were doing 10 or 20 years ago.

NITIN DAHAD

Now, your focus is on Arm, because you’ve got that expertise. Are customers coming to you and taking their own Arm license, or are selling the products based on your own Arm license? In other words, how does a customer work with you?

GRAHAM CURREN

So that that’s an interesting question. Because we’ve had this debate for many, many years as to whether or not when we were only offering design services, whether it was better to buy just the design services from us and the bits from other people. Or whether it is better to supply the whole thing.

What our customers told us repeatedly, what the market told us repeatedly was they wanted a one stop shop. So today that’s what we offer. They come to us and we do everything from, in some cases we write their specification for them, we do their architectural design, we procure all the IP that’s needed, whoever it’s from. And then we procure the silicon, and we test it and we ship it and we ship qualified parts. So we do the whole thing. So they come to us as a one stop shop.

NITIN DAHAD

So they’re not having to do that licensing or anything like that?

GRAHAM CURREN

Yeah, so they don’t do anything. Again, this comes back to what I said right at the very beginning, which is generating that economy of scale. If you go to an IP vendor and you buy something once you’re not going to get the same relationship as you do if you do it over and over again, and it’s not just about price. I think people who don’t do a lot of chips, or maybe have done one, but it was five or six years ago, underestimate how complicated it is to procure, qualify, check and integrate third party IP.

It’s never 100% perfect. It always needs something. It always needs checking. It always needs integrating and so on. So there’s a lot of work in that, and we think it’s far, far better that it’s all done by one and that’s what our customers like.

If I just add one more point to that as well. It’s also a reason why we have never gone into designing our own IP, because when people come to us they say, you know I want to do an architecture and what’s the best solution for me and we need to be independent of that.

So we’re not going to tell them, you’ve got to use this DDR 5 because we’ve designed it, or you’ve got to use this processor because we own the IP for it, we’re not going to do that. We’re going to find the best solution for them.

NITIN DAHAD

Let’s move to some of the current trends. Integration is the big thing right now you know, providing more of a reference solution, that’s what the chip companies are doing. What position does that put you in?

GRAHAM CURREN

Again, it’s very much about solving somebody problem, and in a way I see us a bit as this: we understand silicon and what can be done from it, and our customers typically understand the market and what they need. By putting those two together, what we try to do is to say to customers is “This is what you can do”. And they say this is what I’d like to do, and sometimes of course it directs them in a certain way and sometimes it can be quite enlightening for them. I mean, none of us knew we needed an iPhone until they came on the market until Apple or whoever showed us what smartphones were like and then we all decided. So somebody has to show the customers what’s possible, and what the tradeoffs are. Many of the customers that come to us have not done a design for a while and you know we’ve all seen the exponential curves of cost. They get often somewhat taken by surprise. People don’t necessarily expect a 10- or a 20-million dollar IP cost to be built into the projects so we have to work and show them what’s possible, but also show them how they can retarget a product or change the product or change the architecture to give them the real return on the investment that they need.

There are so many points to consider, whether it’s time to market, whether it’s power, whether it’s unit cost or whatever it is, so we need to look at that, so understanding their business is where we really try to start.

NITIN DAHAD

What do you see as the next big challenge or challenges for the chip industry?  Where do you see yourself having to address solutions in the next year 2, 3, 5 years?

GRAHAM CURREN

It’s very much an evolution of what we’ve been doing over the over the last 20 years, but it’s all about size. To me it’s all about size of data, size of design team.

Again it comes back to those curves that we’ve seen, which are exponential, but it’s easy to look at an exponential curve and think, oh yes, it’s an exponential curve. When you actually look at it and you look at, say, the IBS data and they’re saying – which you can argue with or not – it might cost you 500, 700, 100 million dollars to design a chip. Look at it another way, that’s 5000-man years. And you’ve still got to get it out in 18 months. That’s a lot of people.

Now of course, reusability and all sorts of other things can reduce that. But the steps between are not linear, that even a percentage change, generation to generation, is going up. So when you go from two people to three people, it’s manageable. When you go to from 50 people to 100 people, suddenly that’s not manageable for many companies.

So, we’re looking at big data farms, massive amounts of CPU, massive amounts of disk, huge amounts of data to manage and control and qualify, and big teams. So it’s a market over the last 20 years that has switched very much from being a “I need a design expert” to being, “I need to be able to control everything and manage the project and bring it all together”. It’s much, much more of a large engineering project than it was 20 years ago.

NITIN DAHAD

So Sondrel would continue to be your large engineering team that you don’t need to hire?

GRAHAM CURREN

Exactly, and we need to keep expanding and that raises challenges in the market today. And there’s particular challenges around the market today as we know. Customers are and worrying and the press are worrying about the ability to source the silicon and to get the chips. But it’s equally true in the ASIC and the design services world: lead times are going out, the time from people making decisions and the time when you can even start is going out. There’s a queue of designs starting to build. We’ll see how that pans out in the coming months as well, and of course, fewer and fewer people who are able to do the chips.

And there’s plenty of people who can do your 40-nanometer chip. As I say, there’s only perhaps 3, maybe 4 companies in the world who can do a 5-nanometer chip.

NITIN DAHAD

That leads onto a question which I guess is skills.  I think you deploy teams around the world and I think you’re also sort of hiring in places like India, but is skills a big issue when you’re hiring?

GRAHAM CURREN

Skills is a big issue. The hiring managers often say I want somebody who’s already done a 5-nanometer large SoC. Well, of course there’s nobody, so you can’t hire that, so we spend a lot of effort, and we’ve recently been hiring more people to actually just lead our teaching effort, we worked in the past with universities on courses, we did a lot of work with Nottingham University a few years ago. Training people, teaching them what to do is a really, really important part of our business today because we need to be able to access these teams and we need to be able to scale them up.

And the problem with being at the leading edge is you can’t get them from anyone else, you’ve got to do it yourself.

NITIN DAHAD

I guess you’re you’re hiring them from all over, so you’re distributed around the world?

GRAHAM CURREN

Yes, so our centers are set up to get the right balance between cost access, ability of staff, ability to hire staff, and proximity to our customers. We have to align not only with their cost requirements, but also with their cultural and language and time zone requirements. Because we have to work very closely with them. So we have people in the (United) States, we have people in the UK, Africa, India and China. So we cover the whole world from our different bases.

NITIN DAHAD

Well, Graham, thank you very much, it’s been good chatting to you again.

GRAHAM CURREN

Yes, it’s nice to, as I said at the beginning, nice to see you again and, it will be really nice when we all manage to get out and meet in person a little bit more, as is starting to happen in some parts of the world.

NITIN DAHAD

That brings us to the end of this episode. That was embedded edge with Nitin, and I’m Nitin Dahad. Thanks for listening and see you next time.

embedded edge with Nitin is brought to you by Aspencore Media. The host is Nitin Dahad and the producer is James Ede.