No Time Like MEMS Time
Q&A with Rajesh Vashist, CEO SiTime
This post was made in collaboration with SiTime, who provided educational resources and compensated me for my time. All writing and opinions are my own.
The next generation of high performance processors are really kicking it up a notch. We’ve moved from 300 W to 600 W and now we’re looking at 1200 W+ designs quickly coming down the pipe. TSMC published research a couple of years ago about how to cool these monsters up to 2600 W, and IBM has a lot of patents on intra-die-stacked cooling from the mid-2000s. Moving to larger silicon, or lots of chiplet silicon, at high power is driving high frequencies, and one of the things we don't often think about is how do we get those frequencies - how do we enable the power efficiency of that chip through a smart timing solution that remains flexible and consistent?
Historically, we've enabled this sort of timing through oscillating quartz crystals. These are physically manufactured things, literally with saws and polishing, to within 0.1 Hz, just to make sure that our chips run at the right speed. This industry is being innovated rapidly through the use of in-silicon devices called MEMS: Micro-Electro-Mechanical Systems. In short, instead of vibrating piezoelectric crystals, we can use silicon itself for that function.
The beauty of using MEMS over quartz is multi-faceted. We can enable much more precise silicon timing and oscillators. They consume less power, and suffer lower delay in rapidly changing frequency environments. They can be made programmable, allowing a design that can be applied to a large array of industries with a single unified solution. Of course, there is still a market for custom designs – but the whole point about programmability is that it enables that customization straight out of the fab.
The market leader in this field is SiTime. A couple of months ago we published a brief look into the company, the history, the portfolio, and the market. It funnily enough coincided with spotting one of their chips inside NVIDIA’s Spectrum-X high-performance Ethernet switch at Computex. On the back of that article, I spent some time with CEO Rajesh Vashist, a veteran of the semiconductor space, to find out about their key targets given the recent boom in silicon design.
Ian: Who is Rajesh Vashist?
Rajesh: I've been in the US for 43 years - I've been in the Valley for 40 years. I've been working in semiconductors for 40 years! I've been a CEO for 23 years, and this is the second company I‘ve taken public, I love doing this.
Ian: Is your background in semiconductors?
Rajesh: My background is in engineering from India. The years that I studied were in the late 1970s to the early 1980s. It was a very broad physics, chemistry, and math kind of engineering. So very broad, it allowed us to pick up many different parts of that, and I was very grateful to that.
I came to the US almost immediately after it, because it was very clear that I wanted to do an MBA in the United States. I got lucky - I got financial aid at Marquette University in Milwaukee, Wisconsin. It's a private Jesuit school.
Then I arrived here in San Francisco in 1984 when one of the many booms in Silicon Valley was going on. I started with my first product, which was 64K DRAM, and it was the first product that Samsung semiconductor started with - I was a product manager for it! They were selling for $11 each, and one year later when semiconductors had turned, they were selling it for $1 each. I've been through [commodity markets like] this a few times.
Ian: What brought you to the timing industry?
Rajesh: Well, I didn't actually come to the timing industry per se.
My thesis is that great companies are built when existing large markets are disrupted by a completely innovative technologies. We can see this with Uber, taking 150 year old urban transportation technology. We can see this with Amazon disrupting retail, which has been around for as long as humans have been around. You can even see it with Starbucks with coffee, disrupting hundreds of years of business.
So I saw a timing business, which at that time was $5 billion [a year]. It's $10 billion now, and in a decade it'll probably be double that, maybe $20bn. I saw this market of $5-6 billion was being serviced for 70 years by non-semiconductor quartz crystal companies, which had done a fabulous job, except it's not semiconductors.
So I thought that this was a great opportunity to bring a new technology, semiconductors, into this and bring the value proposition of semiconductors, which is greater performance, programmability, quality, and reliability. There was not much capital expenditure going into it, and that's how I came into this market. I also liked the fact that I thought I could get high gross margins, high growth, and high net income out of it. And 15/16 years later, it's proved that way.
Ian: You expect it still to trend that way?
Rajesh: I do, because as I mentioned, timing has been constant (heh) as far back as the mid-1940s. So in the world of timing, we only see growth. I see more AI, I see more automated driving, I see more personal devices like watches, and I see faster communications. All of them need high-performance under tough environmental conditions, and that's what semiconductor based technology does, That's what we do.
Ian: I think it's important to understand what does ‘high-performance’ mean in timing. I don't need a high-performance clock on my wall, to the nearest second is fine. But in electronics, we need to the nearest nanosecond, right? So what does silicon bring to that equation that quartz doesn't?
Rajesh: I could tell you about 20 different parameters, and I'll rattle some of them off just to show how some of them are esoteric.
One is stability, which is the frequency that's generated by the clocking part or the timing part. Customers care about the degree of fidelity, which is measured in parts per million (PPM), parts per billion (PPB), or now, parts per trillion (PPT). At parts per trillion, you're at an atomic level of accuracy. There is phase noise, in this case less is better. There is jitter - less is better. There's Allan deviation - less is better. There is size, there is low power.
There are even more holdovers and so on, but that must function in the real world. It must function with shock, vibration, temperature variations, airflow. That's what we do - we try to make them resilient. Yet we have made, from the time I started in the company to now, six orders of magnitude better on multiple fronts, such as stability, which is a key metric.
Ian: Six orders of magnitude on stability. That's great for silicon, has the quartz industry from your perspective ever gone forward with those sorts of gains?
Rajesh: They have, but over a 70 year period. The people who invented the technology in the 1940’s would be shocked at how far quartz has come. After all, we put men on the moon with that technology. So it's a perfectly fine technology, except it's not ready for the world of semiconductors. It's not ready for the world of automated driving, It's not ready for AI at the levels that we are talking about.
[Quartz] could do the performance, but not the environmental. Or it can do the environmental, or not - bits and pieces of the performance. So SiTime’s job has been to identify the low-hanging fruit in this large $10 billion market so that we can disrupt it.
We're a very small company - we're estimated by analysts this year to be in the sub $200 million range, which is roughly 2% of the market. Except I believe it's the juicy and the fast-growing part of the market. That's what we're focused on.
Ian: But quartz is fairly cheap, and semiconductors are not. We keep hearing about the latest and greatest being more and more expensive. Obviously, the MEMS you do isn't on those leading edge nodes. But you still have to go get masks, use a process node, build a wafer etc.
Rajesh: When you combine the performance with the environmental stability of our product, I guarantee we’re the cheapest part for that combined. That's like basically saying that in the smartphone market, Apple has say ~15% of the market. But there are perfectly good phones made by Samsung, Oppo, Huawei, and others which are consumed by 85% of the world. We have that iPhone strategy for those who want the high-performance, which is implied in the iPhone. We are that. There are perfectly good phones that do a wonderful job. Most people may not want to use us, but I'm okay with it because on the back of the iPhone, Apple has built a $3 trillion market value company, that sounds pretty good to me.
Ian: Those sound like very lofty goals. Be Apple, be a $3 trillion company!
Rajesh: (laughs) Of course! It’s important to reflect though on the semiconductor market. Recently, NVIDIA crossed the $3 trillion market cap. I'm positive that if 10 years ago we had predicted that, it would have caused tremendous laughter. But it’s not only them - there's TSMC in the almost trillion dollar club, and when I looked recently, I saw Broadcom tending to get to closer to that as well. So this is a new brave new world in semiconductors. This is truly their golden age.
Ian: Some of those companies you've mentioned are your customers! But everybody is familiar with quartz, everybody knows what it does. As you said, it's been around for 70 years, and so everyone seems happy using it. How has that journey been to teach those customers that you have the right solution for them?
Rajesh: So that's a great question.
I arrived at SiTime in 2007, and in that pre-2013 period, our products were not superior to quartz. They were equivalent, and we were living on higher quality, higher reliability, and a better supply chain. We did okay - we got to around $20 million of revenue, not that great.
In 2013/2014, we crossed over, and since then, every product we have brought out has been either remarkably better than quartz or significantly better than quartz. We still maintain our ability in supply chain, better quality, reliability, programmability, and capacity. For example, it was shown with COVID and the supply chain constraints. But we have now gone past them on performance as well. It took us a little while - seven or eight years - but we finally got there.
My thesis is very simple. We're blessed with a very large market, a $10 billion market, In a $10 billion market, we should have such a massively innovative product that customers, when they get it, when they see it, when they observe it, they ‘get it’. The great thing about semiconductors is you don't have to wave your arms around for performance. Customers can plug it in and immediately see if you're talking sense (or nonsense). Hopefully if you're talking sense, they can decide to use it. My belief is that when we make great products, great products are great because they sell themselves. If we have to spend a long time in convincing customers, they're the wrong customers. SiTime is blessed with a large market. We should go somewhere else and sell it.
Ian: You often describe your position of being a market leader in MEMS timing, ahead of the competition because you have a chronological engineering lead. Does the fact that you're getting responses from these high-profile companies mean that you're expecting fast followers to come and be your competition?
Rajesh: Well, here's the thing. Five years ago when we went public, I was asked the same thing by investors. I said that we have a two to three year lead then, and that was five years ago. So it seems to be a perennial two to three year lead, which is already a lot more than two to three years. I think that we are widening the gap because it's not just about MEMS.
SiTime is a timing company - our name means “Si” the symbol for silicon, and time. We're the only company in the world that focused on timing. Nobody else has decided that they should deliver only timing products, whether they are chips, resonators, clocking, oscillators, modules or software. It's only us at SiTime that makes it 100% of our business, and five years from now, we're still going to be focused on timing.
The benefit of this is that we compete with competitors who are 2% or 4% into timing in their whole business. So at our rate of innovation, I am positive it’s going to change the game. We have told the world that we're going to spend something like $70/80 million of R&D. I've been doing that every year for the last few years. So that's our biggest bulwark. That's our biggest moat. Having said that, if and when competitors come, I'd welcome them because it'd be nice to have some competition, I am sure that there's plenty of room if we are 90% market share in the MEMS timing market and that declines to 70%, that’s still pretty decent I would say.
Ian: By that time, you expect the market to be bigger. So does that mean that you end up being an acquisition target?
Rajesh: Not at all. We were acquired once. We were acquired in 2014 by a Japanese company, Megachips. Megachips still owns about 20~% of the company, and in 2019, we spun ourselves out of them. So I'm pretty sure that the best course of action for SiTime is to be in an independent company. I say that because we're the ones who know the most about timing, and there's nobody who can acquire us and help us any more than what we can do for ourselves. So I rather like being independent right now.
Ian: So right now you said it’s a $10 billion addressable market, hopefully moving to $20 billion within a decade. Your three main segments are resonators, clocks, and oscillators. My background is not electrical engineering, so what is the difference between those three?
Rajesh: There are two components that make up the third component. In other words, resonators are exactly what they sound like - they produce a resonant frequency that is consumed with what is called a clock circuit, which is a semiconductor piece, and the clock circuit takes in this frequency and distributes it to one chip or multiple chips. Those are the two elements.
The combination of these two is called the oscillator. The oscillator is the system play. It's the closest to being a fully formed timing subsystem, as it were. So it’s no surprise that SiTime wants to sell the oscillator first. So 90-95% of our product revenue today comes from oscillators, we have been very good at selling oscillators. We recently acquired a clocking company, Aura, and we are selling their products and we expect to sell a lot more of them in the future.
Recently, we have started to pay attention to the resonator by itself and the properties that allow us to sell resonators on their own. There's a lot of physics we can go into, but selling resonators on their own has many challenges. One is they're very, very small. We're talking 0.5x0.5mm. Then there's many other issues in getting them to customers. So we have reserved it for last. Stay tuned in the future for more on that!
Ian: And this is actual physical products, not just IP?
Rajesh: Absolutely. We love selling physical products. We have shipped 3 billion units of these, and we expect to sell billions more.
Ian: I think you've got a few to show me. A member of your team brought these out.
Rajesh: So that is one of our MEMS wafers. This wafer has been processed at our partner fab, Bosch, in Reutlingen, Germany, which is a small town near Stuttgart. You may not be able to see it, but it has a ~100,000 die on a single wafer. This is a display wafer - these are non-functional, otherwise it'd be worth a lot of money! It is our job to test every one of those, mark the ones that are working for yield, and then use them in combination with the aforementioned clocking circuits.
Ian: I guess the yielding characteristics of a MEMS device is going to be different from a traditional semiconductor device - the way you test, the way you dice. 100,000 chips on that 8” wafer, I can't imagine just how complex the dicing is for something that small!
Rajesh: That’s the most fascinating thing. It’s hard, and probably the biggest difference between us and regular semiconductors. In this technology there are moving parts inside each chip. It's a near perfect vacuum and there's a MEMS on the inside. It’s on a physical anchor and it is vibrating millions of times a second, under all kinds of temperature and pressure.
Ian: You give it a voltage to do that.
Rajesh: Of course, we have to start the vibration, as it were. So there are features we have to worry that most semiconductor companies don't worry about - for example, the elasticity of the device. Does it ever break? Does it fatigue? It's a material after all. It's a moving part.
Ian: There was a story a while ago with Tesla’s AI chip - they had to deliver power at a thousand amps per square centimetre. The timing oscillators nearby were experiencing failures at random times because the oscillators that they were using didn't have the right thermal characteristics and were breaking.
Rajesh: Those are big chips, the ones that you're talking about from Tesla - but we find that that even sometimes in a watch, sits an oscillator, whether it's ours or somebody else's, next to a power management chip the spike in temperature in milliseconds can be as high as 150c. That's amazing, right? So that jolts the material out of specification, and to keep on functioning with that and shock and vibration, that's why we spent all that money on R&D.
Ian: But quartz doesn't have to worry much about that? The comparison is that it’s bigger, it's not as accurate, and it consumes more power?
Rajesh: Well, it depends on which one we're talking about. In this case, they could do that, but they're probably going to get big excursions in the frequency with temperatures.
Ian: Oh, you're talking about drift?
Rajesh: Drift and just thermal gradients. Everything always existed before SiTime came to the market - it's just like coffee existed before Starbucks came, right? Or retail existed before Amazon came. Amazon just made it better. In this case, SiTime just makes it better.
Ian: A big draw in new AI hardware right now is the efficiency of the chip - how much power can we give the chip for compute within a fixed power budget. How much pickup are you seeing across that market, or at least the trend towards MEMS timing?
Rajesh: We see a trend towards precision timing for sure being used more and more in all parts of AI. It is in the training, this is in the NICs, it is in the accelerator cards. It is in the switches, it is in the interconnects, it's in the optical interconnects, it's in the cabling. So we see quite a pickup, and we have guided our investors that this year as a consequence. So we see significant growth in SiTime as a result. We started selling dollar priced chips, and we're moving now slowly up the chain to 5, 8, 10, 12, and 15 dollar chips as we go forward. And I don't think its stopping. I think it's going to accelerate towards more consumption of more timing.
As an example - optical cables have always been used to interconnect except they have been passive cables. They're not ‘smart’ cables. In AI, because the throughput is so high, the latencies need to be very low. In other words, signals need to get there fast. For a variety of other reasons, some AI builds have moved from passive cabling to active cabling. That was a thing that didn't happen very much before.
So there are trends like this. Because as you said, there's literally trillions of dollars being put into AI. Every ounce, every micro-ounce of performance that could be eked out is going to be eked out. That's where SiTime shines - where people care about performance in tough environmental conditions.
Ian: So you say you're investing into the R&D, This $80 million number that you gave earlier - when you say R&D for a company like yours, what does that mean? What are you creating with the R&D budget?
Rajesh: Well, first of all, when we look at technical companies, SiTime is one of the deepest technology companies I've worked for, and I've been working for 40 years. That's what keeps me going.
We have MEMS technology - we are innovating there.
We are innovating in our CMOS technology.
We are innovating on putting these together physically.
We are innovating in the packaging of this and,
We are innovating in how we sell.
We are about 400 people, and I would say 300 or so people are all engineers – that’s whether they are working as engineers or not. I'm an engineer, our salespeople are engineers, our marketing people are engineers, our business people are all engineers. So what we are basically saying is that we are about spread out.
When I came to this company, our innovation was only in the MEMS. Then we moved additionally to the CMOS piece. Then we moved it to the systems, then we moved it to packaging. So we have found an enormous place for innovation. Recently, we have been spending time and effort on synchronization software, which is exactly what it sounds like. It is software that synchronizes different elements of this timing subsystem. So it's always accurate, and deals with latency and those kinds of concerns.
So our scope for innovation is tremendous, and we can't find enough people here. We are in Silicon Valley, we can't find enough people. So that's why we're in Michigan, we're in Netherlands, we're in Japan, we're in Finland, we're in Ukraine, we're in Malaysia, we're in Taiwan, and we're recently in Bangalore, India. So really we spend a lot of time innovating, finding engineers where they are.
Ian: So you're hiring?
Rajesh: And we're hiring!
Ian: It's interesting you brought up packaging, because one of the other things I wanted to talk about is this OXCO comparison. OXCO enables thermal stability, and these devices are larger than what you showed me with the wafer. Explain to me what I'm seeing.
Rajesh: The big component is a very high performance crystal based oscillator, a quartz crystal. To the left of that, the much smaller one, is SiTime’s module at equivalent or better in performance.
We offer a much smaller volume, with 2x better performance and 1/3rd the power. Our product is called ‘Epoch’, which we've been sampling for about a year and selling for the past five to six months. Epoch sells for anywhere from $20 to $30, so it's an expensive part by comparison, but it's truly a module in that package. Inside it are multiple chips, multiple semiconductor analog circuit chips, and multiple MEMS chips.
So to your point on packaging, we spent two years developing this highly thermally resistant package. In fact we actually did have to do multiple simulations of that, and we have some experts in the company who are experts just in doing thermal simulations of things like this.
Ian: In order to get parts per trillion (PPT) timing accuracy, you need a thermally stable environment, and this is that product?
Rajesh: This is that product. But you still intrinsically need stability in your MEMS. You also need stability in the analog domain, and you have to do the thermal compensation so that in the presence of very high temperature variations, we can compensate quickly and accurately. Both are important - it has to be quick, and it has to be accurate, and SiTime has that. We have more than 100 patents, and several others that have been filed, but we also have technical knowhow, which is almost as important because that's kind of the dirt under our fingernails.
Ian: You’ve characterized a lot between compute and networking - what about when it comes to wireless and telecommunications? We’re a good way into the 5G installation cycle, and 6G is being discussed. Is that going to be a big win for you?
Rajesh: We hope so. One of our business units is focused on communications, enterprise, and data center. Right now, it's the time for data center and AI to shine. But very quickly to your point we expect the communications piece and the enterprise piece to start shining.
But we also do business in the consumer end. We are in some of the more popular smartwatches, we are in some of the smart doorbells that you find. We are also in some electronic cigarettes for some reason! We are in many other interesting things.
We’re also in aerospace and equipment, we're in functional logistics. We are in high-speed communication - because as you can imagine there are standards put out by the US government for highly encrypted, very non-spoof-able or non-disturb-able communications on the field, so we are part of those.
We are in many cars, particularly US electric cars, as well as Chinese electric cars. We have around 300 headline use cases for our chips, and in that we have identified 10%, 30 or so, that are growing very, very fast. Our job is to innovate along those lines and keep the company growing at about 30% a year.
Ian: One of the phrases I like to quote is that the ‘people who make the pickaxes never know what the pickaxes are going to be used for’. So what's perhaps one of the more interesting or esoteric or unexpected use cases?
Rajesh: Well, I would say it has to be the e-cigarette. I never thought that pulling tobacco and burning it at that precise temperature required very precise timing, but I was blown away when I learned that. It's a very famous brand.
Ian: Is that the gamification of that market?
Rajesh: To be honest I don't know - I'm not a smoker myself, so I wouldn't know. But obviously it's doing well for people.
By contrast, we are in other things on a healthier note. Sleep indicator rings, oximeters, and glucose monitors - you know, healthcare. I like that field a lot! It's a little bit conservative, slow to adopt, but I think they're applying SiTime products there because they bring tremendous value; and I like that story too.
Ian: What you think the audience should take away from this? SiTime isn’t a household name.
Rajesh: We're a company that is disrupting a market that nobody's heard of: timing.
Both you and I, we admitted that we didn’t know much about the timing market before engaging in this space. Before I came, I didn't know that timing was this big, and you admitted the same. So for timing, SiTime is disrupting a relatively unknown market. We’re the absolute technical leader in it, and we've been doing this for a long time.
We're also a highly profitable company, and we have a huge market potential which we've just begun to penetrate. So I'm very excited. We went public in November 2019 at a market value of around $300 million. Recently that has reached $3 billion. It's done rather well in less than five years to grow 10x. I'm very proud of that - but we've also solved some of the world's very urgent problems.
I think the team inside the company takes a great pride in solving customer’s tough problems, especially those that change the world - like satellites and AI and others.
This was really good. Thank you.