It's About Time

I recently gave a talk at the Yole MEMS CTO Meeting. This was an invitational gathering of about sixty MEMS CTOs and equivalents to discuss our profession. The meeting was organized by Jeff Perkins of Yole and Alissa Fitzgerald of Fitzgerald Associates. I chose to talk about the CTO job function, and we scheduled my talk to be the opening presentation in order to set the tone for the meeting.

In the meeting I described what MEMS CTOs do, what sets us apart from other technical specialists, about our responsibilities to our companies, about our responsibilities in general, and particularly about our responsibilities to Humanity. I don’t think many of you are MEMS CTOs, so I will skip the technological and job function parts. The important part is our responsibility to Humanity.

When all is considered, that is our most important job.  It is okay to work for money (we need money to get by).  It is good to work for enjoyment (we should definitely enjoy what we do).  But it is vital that we work toward the betterment of Humanity.  I use that term rather than others, for instance ‘making the world a better place’, because Humanity denotes both the small and the large, in contrast to talking about the world as a whole which can seem overwhelming.  We can do our jobs, however small or large, and still contribute to Humanity.

Engineers and technologists do make a contribution. For instance in the MEMS field we build accelerometers and gyroscopes that detect automotive impacts and impending spins to deploy airbags and stabilize braking. These sensors have saved at least 100,000 lives in the last decade. This is a contribution to Humanity. A doctor that prescribes a life saving drug sees his or her contribution to Humanity directly, but the engineer that helps design the equipment that makes that drug has a less tangible but vital contribution.

How does this apply to SiTime? We make millions of precision timing chips each month that are built into a wide range of products. Each of these products contribute; some more and some less. Integrated over the whole, these products are tremendously important. They spread joy, they improve communications, they save energy, and some indirectly save lives. Millions of times a month SiTime contributes to Humanity.

I recently served on an expert panel at the 2010 International Frequency Control Symposium. The IFCS is where the world’s leading timing and frequency control scientists meet to discuss the latest advances. Our panel was formed to debate and answer the question, “Will MEMS Replace Conventional Quartz Resonators?”

The panel was convened by John Vig, a past president of IEEE, and included past and present DARPA program managers, a leading expert on telecom filters, top technologists from the two largest quartz manufacturers, and myself representing the MEMS timing industry. The audience included over two hundred timing experts ready to cross examine the panel and dissect every response over an intensive two hour debate.

These panels are serious business. They are where scientific consensus is forged and from where industry and government is moved. Some panels are more important than others, and this one had shaped up to be very important. Scientists in the field described it as “The Famous Panel”. Ultimately, what was being evaluated was the future of a critical technology and a multi-billion dollar industry. Are things normal, or is the ground shifting?  Will the next ten years be like the last forty, or is a radical change coming?

I obviously have a position on this. The question is not “Will MEMS will replace quartz?” Across computers, scanners, printers, cameras, video equipment, networking equipment, and dozens of other applications, MEMS is already replacing quartz. The question is “How fast and extensive with the change be?”

I was ready to point out that the transition had started, but didn’t need to make this point at all. In his introduction John Vig refocused the debate, pointing out that the change was already underway and that the debate would instead center on how the change would unfold. Then surprisingly to me there was zero dissent. Everyone on the panel agreed! MEMS was already replacing quartz and will make further and deeper inroads. Even the two experts from the world’s largest quartz suppliers agreed.

The debate moved forward to define how extensive the change will be. Can MEMS provide the frequency stability? YES, a paper the previous morning from Professor Thomas Kenny of Stanford showed OCXO stability. Can MEMS provide low enough phase noise? YES, a host of MEMS papers at the conference described impressively low phase noise. Can MEMS be made inexpensively enough for consumer applications? YES, SiTime has gone toe-to-toe with quartz and won. Can MEMS provide reliability and quality? YES and YES, semiconductor fabs are known to provide better production control and higher quality than quartz fabs, and this control and quality is inherited by Silicon MEMS made in semiconductor fabs.

The question of MEMS replacing quartz was decided and considered obvious. The speed of the change will be fast and the extent deep. “The Famous Panel” and the audience of experts debated and passed their judgment. MEMS is replacing quartz, it is a done deal.

May was another great month at SiTime.  We shipped our 20 millionth oscillator, our customer base grew, our production volume grew, and the number of end-user products with our parts grew.  Every quarter for the last year and half has been bigger than the quarter before.  In the middle of May, just half way through Q2, we surpassed our Q1.  We are also seeing strong design wins and pre-production orders, so we feel good about the rest of the year.

What is behind our growth?  First, we have a happy, loyal, and growing customer base.  Our existing customers are using our parts in more of their product lines while new customers are designing our parts in.  Second, we have an extensive product portfolio.  We have clock generators for multi-frequency generation, differential oscillators for high speed digital systems, low power oscillators for handheld applications, spread spectrum oscillators for reduced EMI, and various other specialized product.   And of course we have general purpose oscillators too; our new SiT8103 is a great oscillator for general functions, with low jitter, moderate power, and a wide frequency range.  Third, the quartz industry is doing themselves no favors.  One of our large customers, one of the world’s largest buyers of quartz oscillators, is seeing a twelve week lead time for quartz.  Can you imagine that?

It takes months to get quartz oscillators, but only weeks to get SiTime’s semiconductor-based MEMS oscillators.  Why the difference?  Because we rely on standard semiconductor production processes, not specialized hand-built processes.  We are ramping our production quantities but that is not causing delays because the semiconductor industry knows how to ramp.  Does it all the time.

A few days ago the March/April IEEE Micro magazine arrived at my desk. This issue is dedicated to a few selected papers from last summer’s Hot Chips meeting. Out of 75 submissions to Hot Chips, 27 were accepted for oral presentations, and then only 7 of those were selected for publication in IEEE Micro. These papers are all excellent, and one of them is written by Dr. Sassan Tabatabaei of SiTime. It is posted here on SiTime’s web site.

High speed digital systems particularly benefit from SiTime’s MEMS-based silicon oscillators. One can divide digital applications into three types: (1) state machines, (2) serial and parallel interfaces that include a clock, and (3) serial interfaces that embed the clock with the data. To be brief I will address only the state machine case, which includes microprocessors and FPGAs.

The ideal reference clock for a state machine must be high frequency and have particularly low jitter. High reference clock frequencies are beneficial for state machines because they  simplify the work of their internal PLLs. All fast state machines rely on internal PLLs to multiply their reference clocks to their internal clock rates and produce the various phases and clock domains used to sequence their computations and data flow (for example at 2.345 GHz). The first problem they have is they must usually multiply up rather low reference frequencies (like 27 MHz) to their multi-GHz internal frequencies. The greater the multiplication, the worse the clock jitter. And to make it more difficult, in order to match external data requirements they must often multiply by highly non-integer values (2345/27 = 86.8519 in the example) which requires pre-dividers that reduce clock quality further. Finally, these PLLs are analog components that must work in very noisy and harsh environments. This all produces poor internal clocks that reduce timing margins and close data eyes.

By providing higher clock frequencies we simplify the task for these PLLs. Less multiplication gives less clock jitter. If special output frequencies are needed then SiTime’s clocks can readily be programmed to deliver integer divisions of those frequencies (for example 234.5 MHz), so the state machine’s PLLs have an easy integer job (in the example just 10x) and no pre-divides. Before SiTime one could not readily do this – most high frequency oscillators were SAW or overtone devices, and you couldn’t just order them up in any frequencies you wanted. The result with SiTime oscillators – one gets a lower jitter internal clock with improved timing margin and wider data eyes.

This is a very big deal for cutting-edge digital applications. An executive with a very high-end FPGA company came to me after our Hot Chips talk and said, “Thank you, you are giving us improved timing margin and we are going to suggest our customers use your oscillators.”  Think about that! This company makes some of the world’s fastest chips and their preferred clocks are from SiTime.

Two thousand and nine was the year MEMS-based semiconductor oscillators became ubiquitous.  SiTime has shipped oscillators since 2007, but in 2009 we saw significantly larger volumes and wider applications.  Our sales increased quarter over quarter and our customer base steadily expanded.  But something else, something unexpected for me, happened last year.  We developed a large enough application range that I could no longer say where our oscillators were going.  While I’m sure our company databases have the information, it is more than a single person can follow.

I know we have millions of parts in flat panel televisions, I know we have parts in laptop PCs, and in networking equipment, cameras, phones, printers, set top boxes, disk drives, and many other applications.  But some of our customers are expanding their usage without cross-checking with us.  Our parts are simply working for them and they are building them into additional products.  One of our customers is ordering millions of parts for consumer applications.  I don’t know exactly for what, but they are happy with our parts, our delivery, and our quality.  I know this because they told us so, and they are steadily increasing their order rates.

That is what happened in 2009 – SiTime’s MEMS-based semiconductor oscillators began to significantly replace the legacy technology across a wide range of products.  It is likely as you read this post that you have used our oscillators.  Perhaps in your computer showing this text, the servers that sent the data, the network equipment that carried the data, the destination monitors in the airport through which you last flew, the phone on your desk.  Who knows, but millions of our parts are running in products around the world, and you may already be using them.

The quartz industry has a reputation for poor quality.  Some of it is because it is difficult to maintain consistency in quartz production.  However, part of it is self inflicted by the way quartz companies resell and relabel their product.

Because there are so many combinations of frequency, package, supply voltage, and various other specifications, there are thousands of “standard” quartz resonators.  Far too many to keep in stock or tool-up to build.  So the quartz industry has developed a resale-and-swap model where many, and in some cases most of the product a supplier may “manufacture” is actually bought from partners or even competitors.  This is how many quartz suppliers offer a wide range of product – they re-label parts from manufacturers.

In general re-labeling is fine and does not cause problems as long as the manufacturers are high quality and the relationships are durable and consistent.  But in this case the agreements are often ad-hoc and temporary.  They are often short term buy-sell deals rather than corporate relationships and there is often a lack of quality control.  Problem is that the customers think they are buying well qualified and stable product.  But really the parts are coming from wide ranges of unqualified manufacturers.  And because the ceramic quartz packages tend to look the same (there are only a few package manufacturers) the suppliers can change manufacturers.  They can and do even change manufacturers in the middle of their customer’s production runs.

Granted the high quality suppliers don’t do this, but many others do.  And chances are you may be buying untraceable product.  If your supplier changes manufacturers or the manufacturers change their process then the parts you buy might not work as you expect.  Specifically, they may not work like the ones you qualified.  They may shift in production; they may fail in your application circuit.  How does quartz fail?  Main problem is latent startup issues.  They may work well enough in your factory, but when they are cold or hot they may not start.  Scary?  You bet!

A few years ago my brother gave me a fine pair of audio speakers and I have recently been looking for a decent amp to drive them.  I want something well built, but not too expensive.  I am not a big electronics consumer and do not even have a TV, but since I am an analog engineer I am particular about amp quality.  That has sent me looking at and listening to amps in audiophile shops.  Most of the amps at these stores are solid state systems, but a few are vacuum tube designs.  The tube amps don’t perform very well, they distort the signals, but some people like “the tube sound”.  I don’t.

That got me thinking about tubes.  When I was a kid there were tubes everywhere.  They didn’t actually work very well, they burned out all the time, they had low gain, they came in only one polarity, they were hot enough to singe my fingers, and the voltages on them could shock me!  I didn’t like tubes, transistors were much better.

Nowadays tubes are used in just a few narrow applications.  High power microwave sources and RF amps still use tubes, and some audiophiles like they way their distortion sounds.  That is about it.  Application by application tubes have given way to smaller, higher performance, more efficient, and more reliable silicon transistors.  That is what is happening to quartz crystals.  Silicon is replacing them – application by application crystal oscillators are giving way to silicon oscillators.

Two years ago SiTime introduced the first commercial MEMS-based silicon timing solutions.  These support processor and general clocking applications and offer programmability, short lead times, and fantastic reliability.  Eighteen months ago we introduced low jitter oscillators for serial datacom applications.  Ten months ago we introduced high frequency oscillators (up to 800 MHz!) for very low jitter differential applications.  We also have spread spectrum and voltage controlled oscillators, and we are now releasing multi-frequency clock generators and low power oscillators to full production.  Our new SiT8003 takes less power than the vast majority of crystal oscillators, and supports battery operated and hand held applications.

So we have general purpose, low jitter, high frequency, spread spectrum, voltage controlled, multi-output, and low power oscillators.  And one by one quartz applications are moving to silicon.  Eventually quartz oscillators will be relegated to a few narrow applications.  Perhaps some audiophiles will like how they sound.

We are in the early stages of a historic change in the timing industry. This change is important and will impact the majority of electronic products. It is worth writing about and worth reading about.

I will discuss this change and the new timing industry it is creating. I will describe how Silicon Timing, with MEMS at its core, is replacing Quartz Timing.  I will describe how it will grow to impact crystals, oscillators, clock generators, clock distribution, and nearly all facets of electronic timing.

Here is the back story: Quartz crystals are used to build precise clock oscillators for electronic systems. They sequence computation and communications in the vast majority of electronic products, with about fourteen billion units sold this year. Quartz is available as resonators and oscillators. When people say “crystals” they are referring to passive resonators that must be electronically driven with external circuitry. When people say “oscillators” they mean active components that include resonators with drive circuitry in one or two packages. Quartz crystals and oscillators were first developed in 1920 and have been slowly and steady improved.

Silicon Timing is now replacing Quartz Timing. In Silicon Timing the precision timing signals are made by silicon resonators and silicon circuits. At the core of Silicon Timing is a technology called MEMS, which is an acronym for Microelectromechanical Systems. With MEMS technology we can build tiny mechanical components that sense motion, pressure, chemicals, or in this case vibrate at precise frequencies. The first MEMS oscillators were described in 1966; it has taken the intervening forty years to develop the technology and bring commercial products to market, but now Silicon Timing is commercially available and servicing a wide range of applications.

SiTime is the first company to produce commercial silicon oscillators and we lead the field. We are a spin-out of Bosch, the highly respected German automotive supplier and one of the world’s largest MEMS companies. Working at Bosch and Stanford University we developed the technology that would become SiTime. We realized that we had the final pieces of the technology necessary to make commercial Silicon Timing possible, and we set up SiTime to do that.

The figure below is an illustration of how new technologies displace old ones. It shows an incumbent with a dominant position and a new arrival growing and displacing the incumbent. The incumbent technology is not completely replaced, but usually survives in narrow applications, while the new technology grows to levels never achieved by the incumbent. We have seen this pattern repeated across every area of technology. This is how tractors replaced horses, airplanes replaced trains, transistors replaced tubes, and email replaced faxes.

This is how Silicon Timing is replacing Quartz Timing. How far along are we? Just starting.