Thank you, Tegus, for sponsoring this post. Also, thank you to anyone I had conversations with as I was getting up to speed on the industry. Lastly, I want to thank you — the reader. Without your support, this newsletter wouldn’t be possible. Every day I’m grateful for that. If you find this post useful, consider subscribing or sharing! Anything helps! Now let’s talk about timing.
Keeping track of time is pretty important, and before my recent dive, I truly had no idea how timekeeping was technically possible. Today’s most popular timer is the quartz oscillator, a time-tested technology that sits in most phones, watches, and computers. A quartz oscillator uses resonance to keep track of time.
Resonance phenomena occur whenever a vibration happens, which can happen mechanically, acoustically, or electromagnetically. The classic example of resonance in timing is a grandfather clock. The pendulum swings back and forth, and the time for the arc time stays constant, which is one way to measure time.
This was further exploited in wristwatches in the mechanical watch with its balance wheel (a smaller pendulum) and eventually in quartz timers. Quartz won out over mechanical designs because it was cheaper, smaller, and more reliable. But unlike mechanical designs, which rely on a constant arc to time, Quartz relies on piezoelectric effects to tell time.
Piezoelectric effects happen when certain materials have an electric charge that is emitted in response to mechanical pressure or heat. So when mechanical pressure is applied to certain quartz crystals, an electromagnetic charge is emitted. Quartz timers rely on the inverse of this effect. A voltage applied to quartz changes its shape, and when removed, it oscillates at a stable frequency. By continuously applying a voltage to quartz, certain quartz crystals will resonate at stable frequencies that can be used to measure time or other applications. The cut of the quartz determines how it will resonate, which is important for different base frequencies for quartz.
Most quartz crystals are cut specifically for the desired frequency — for example, the popular 32.768 kHz frequency. The reasoning behind 32.768 is that it’s 2^15 (32768). Because its base is 2, it’s easily manipulatable by circuitry to count time by dividing 32768 by 2^15 to get a 1Hz signal. The desired end frequency is often multiplied or divided to get the net end goal, like the 1Hz example. Division of frequency is preferable as there is noise, and multiplying noise in the signal amplifies it, while dividing doesn’t. “Division,” in this case, is simply counting and skipping clock cycles to a lower frequency.
Telling time is just one crucial function of an oscillator. Semiconductor devices have to be in sync for the system to function accordingly, but beyond every semiconductor device having some kind of timing device nearby, there’s a whole class of various frequencies from oscillators that are not just used in timing. Oscillators can have many functions beyond just timing itself.
Oscillator frequencies can vary wildly, from hertz to gigahertz applications. The frequencies needed in audio, communications, display, and multiple other devices rely on specific frequency resonators and chips. Examples are modulation schemes for networking, which require frequency transmissions in certain ranges. If everything was at the same frequency, the interference for all these devices would ruin the signal. So specific frequencies are designated for specific applications. I found this Wikipedia article great for a common list of applications.
Just telling time isn’t enough. There are several attributes to what makes a good clock, and I’ll review a few of these attributes and tradeoffs. Namely Q-Factor, jitter, PPM, and temperature range. Timing is technically hard, and external factors can ruin the signal. Each of these factors is about minimizing some external factor.
The q factor is a parameter to measure the energy stored in the resonator versus lost in a cycle of oscillation. Higher Q means lower energy loss and vice versa. This is important for energy. Quartz has a surprisingly high Q factor, often in the 10^4 to 10^6. So the higher Q, the more stable the resonation.
Lower Q means the oscillation dies out quicker over time.
Meanwhile, Jitter focuses on phase noise. Jitter is especially important for telecommunications as it measures the deviations of the signal from the desired signal. Jitter is a major problem in all communication as any noise introduced to the signal decays the signal strength. It can be caused by electromagnetic interference from other frequencies and resonators or random stochastic noise. This is calculated in Root Mean Square value in pico-seconds. Often simply quoted as picoseconds — like .23ps. Think of phase noise as the error that is accumulated over time.
Lastly, PPM or parts per million. PPM is a way to compare the accuracies of different crystals. 100 PPM is 100 / 1,000,000, or 10,000. You can divide this number by the number of seconds in a period, which will be the error rate. A 100 PPM in 1 day (86,400 seconds) would lead to 8.64 seconds per day of loss. A lower PPM means less loss. Comparatively, a typical watch is 20 PPM, which means it loses 4.32 seconds daily. The interesting part is that PPM is not constant over temperature, which is another important factor. The PPM of an oscillator is not constant, and at higher and lower levels creates more loss or gain. PPM is measuring frequency drift in particular.
This is where temperature stability is important. Certain oscillators are more sensitive to ranges in temperature. There are methods to correct these frequency changes in temperature, and that often comes from the analog semiconductor attached to the device. The entire class of Temperature Compensated Oscillators is called TCXO.
Other attributes are important, such as size, price, and voltage, but I’ll be discussing that more in-depth later. Also, if you’ve made it this far, you’re already a champ in my eyes, so let’s keep going.
The primary point of this article is to introduce you to something that is relatively new in the semiconductor world; MEMS. MEMS stands for Micro-Electro-Mechanical Systems and is a class of semiconductor devices that are tiny mechanical systems in silicon. These are tiny gears, sensors, accelerometers, and other mechanical processes that are shrunk to the microscopic level. Remember, physics works the same even at microscopic levels, so an accelerometer can take almost no space if shrank down.
An important emerging class of timing solutions introduces MEMS devices to timing. In this case, instead of using the piezoelectric properties of Quartz, MEMS use the properties of electrostatic resonation and cantilevers to resonate at a tiny scale. Imagine the tiny piece of silicon vibrating like a ruler at the edge of a table; if you apply force, it’ll resonate at a constant frequency.
The benefits of MEMS are obvious and have been sought after for a long time. MEMS is a silicon-based technology, and silicon is one of the cheapest and highest reliability manufacturing techniques available to humans. Importantly MEMS, can be miniaturized and put on the chip, and co-packaged, which is important for smaller devices like AirPods or other smaller electronics.
MEMS is not a perfect technology, but it is silicon-based. That gives a slight cost advantage over the entire Quartz-based complex in the very long run. The playbook of scale manufacturing is proven in silicon and should continue to get cheaper.
However, I want to discuss the primary company offering MEMS timing solutions today, or SiTime. SiTime has 2 core resonators. Their 32-kilohertz resonator, and their 40 MHz resonators. This competes with a litany of specific frequency cuts from quartz, but there are puts and takes there that I will discuss in the MEMS vs. quartz section.
The timing market isn’t all 32-kilohertz products like I mentioned above. To serve the large list of timing-related needs, there are 100s of different SKUs for oscillators and resonators. It’s probably a good time to differentiate between a resonator and an oscillator.
This is a bit tricky. The entire market can be broken up into three types of devices at a high level. Resonators, Oscillators, and Clock ICs.
Resonators are the actual vibrating piece of Quartz, Ceramic, or MEMS. Alone it can be attached to a semiconductor sometimes and compensated usually with an ASIC or MCU. Sometimes semiconductor devices don’t need an oscillator, as the circuitry for the compensation aspect is attached to the microcontroller unit (MCU). That is a class called a passive resonator.
A resonator with quartz or MEMS attached to an ASIC or semiconductor to create a consistent resonating frequency is called an oscillator. For something to oscillate constantly, there needs to be a device that can regulate and continuously give the resonator electricity. The package of a resonator and a chip is an oscillator.
Lastly, there’s a clock IC, a device that gives multiple clock references. For example, FPGAs could have different time domains for the data path, control plane, and memory controller. Clock ICs consist of multiple oscillators or resonators packaged together to tell multiple times.
Important here is telling the differences between the products. Resonators are often components for oscillators, while oscillators often are components for other products such as PLLs. Clock ICs can even contain oscillators, and this distinction often blurs. But for this market sizing conversation, I’ll reference these 3 markets with this nomenclature.
According to SiTime, the total market is approximately 8 billion dollars. If you talk to industry experts, this number can vary widely, but I will stick with the $8 billion dollar number. The composition of that TAM is ~$4 billion in crystal oscillators, ~$1-1.4 billion in Clock ICs, and ~$3 billion in Resonators. Regardless it’s a sizeable market and grows partially with semiconductor volume and value but at a lower rate.
My guess as to longer-term growth rates is in the low single-digit range or approximately 4%. Timer’s greatest torque is the volume of semiconductors, and as the number of devices proliferates, timing volume may be higher. However, I think it’s correct to consider it a low single-digit growth market.
If this couldn’t get any more confusing, I want to now also differentiate between oscillator types. Mainly these are XO, VCXO, TCXO, and OCXO oscillators. These stand for
- Crystal Oscillators, aka Standard (XO)
- Temperature Compensated Oscillators (TCXO)
- Voltage-Controlled Crystal Oscillator (VCXO)
- Oven Controlled Crystal Oscillator (OCXO)
There are other subtypes of oscillators, but I’ll leave that out for this analysis. The simplest and cheapest type is an XO oscillator. These are often the lowest precision and extremely inaccurate in higher or lower temperatures. The ASP and performance on XO oscillators vary considerably. They are not high accuracy (PPM) nor highly temperature stable. However, they can be as cheap as 5 to 10 cents.
The next category is a TCXO or a temperature-controlled crystal oscillator. I know the term crystal shows up, but TCXOs now can also include MEMs. TCXOs have much higher temperature stability. The TCXO contains a XO and a temperature-sensitive correcting device that accounts for temperature changes. The TCXO offers an order of magnitude of temperature stability over a XO. The ASPs for TCXO are often in the ~50 cents to a $1.00+ dollar range.
A VXCO or voltage-controlled crystal oscillator focuses on a device that can be adjusted with an externally applied voltage, hence voltage control. Importantly this is extremely desirable in a PLL (phase locked loop), an important component in networking. They often can have much higher frequencies than XO or TCXO products. The ASP for these products varies from the $3-10 range.
Lastly, we have the OCXO. OCXO or Over Controlled Crystal Oscillators have much higher temperature control rates and high stability over TCXOs. These can be orders of magnitude better than TCXOs and consume high power to keep stable time. OCXOs have a miniature oven to protect the crystal from external temperature changes, which improves stability but creates a cold start problem (it takes time to warm up the OCXO) and costs quite a bit more. OCXOs often cost in the $30-100 dollar range.
A simple rule of thumb is that the higher frequency, precision (PPM), and lower noise (ps), the more expensive the timer is. It’s unsurprising that most of the volume is on the lower end (XOs), while a meaningful amount of the dollar profits for Quartz companies is on the higher end (TCXO+).
The largest market is on the lower end, with XO (standard crystal oscillators) compromising ~60-70% of the market's total value. These can range from 1kHz to multiple MHz, but importantly these are low precision products. The units for this market are in the 10s of billions annually and go into almost every device we use today.
The next step up in the market is the TCXO, VXCO, and OCXO markets. These are much lower volume ~(800-900 million units annually) but have much higher corresponding ASPs. While an order of magnitude less in terms of units, the total value of this market is ~30-40% of the total addressable market.
I want to spend a second talking about MEMS and Quartz. This is a preview of my next post, but I believe that MEMS, as a technology, is improving quickly and will likely take a meaningful share over the coming years. MEMS is not a perfect technology, but because it has a roadmap to improve its quality and a roadmap to improve its price (silicon fabrication), it likely can be a much larger part of the $8 billion dollar market than it is today. For context, MEMS is approximately ~2.8% of the 2021 TAM. Before we talk about that, let’s lay out some major players and the important competitive dynamic in 2020.
The largest players in the world are dominated by the Quartz players. Here’s a quick profile of each.
Seiko Epson (6724:TWO) Seiko Epson is the largest manufacturer of quartz oscillators globally. It’s part of a conglomerate that sells printers, which ironically needs many timing devices. The company is public, and the timing segment is within the semiconductor and microdevices segment, representing 11% of sales.
Nihon Dempa Kogyo aka NDK(6679:TWO) - NDK is a pure-play quartz manufacturer and makes a 31% gross margin and a 12% EBIT margin. Revenue increased 15% YoY last quarter.
Asahi Kesai (3407:TWO) Also known as AKM, this large conglomerate manufactures most quartz ASICs for oscillators. They notoriously had the AKM fire that threw the supply and demand out of balance in November 2020. Quartz is just a small part of their large conglomerate. I cannot parse how large it is within AKM.
KDS Daishinku (6962:TWO) KDS is a pure-play quartz timer company. Here’s a brief history of the company. Their revenue outlook for 2023 is 10% revenue growth, 15% operating income, and improving ROE. They have a 34% gross profit margin.
TXC Corporation (6042:TT) TXC is a pure-play quartz company that took meaningful share this year. They grew revenue ~38% YoY last quarter and boast a 36% gross margin, an industry-leading profitability level for quartz.
Taitien Electronics (8289:TT) is another pure-play quartz company. They have a ~28% gross profit margin.
SiTime - (SITM: Nasdaq) A pure-play semiconductor timing company. They benefitted massively from the AKM fire but has a 90% share in the MEMS market and is the only scale player for MEMS oscillators. Importantly SiTime has a ~65% gross margin.
Microchip (MCHP: Nasdaq) is a large semiconductor conglomerate. I have to mention Microchip because they are the only other MEMS-based timer company after they acquired Microsemi, which acquired Discera. The MEMS business seems to be subscale within Microchip, but if there ever was a competitive push against SiTime, it would come from Microchip.
The 2020 AKM fire was just one of the many disasters of 2020 but probably the most important event for the quartz industry. As I mentioned, an effective oscillator requires both a resonator and circuitry. In the case of quartz, most of these ASICs were fabricated at AKM, which threw the TCXO market into chaos. Shortages in quartz are not new, as the 2011 earthquake in Japan disrupted NDK’s components business meaningfully.
However, this also happened in conjunction with a global semiconductor shortage, and of the many component shortages, such as ABF substrates, timers went to the top of the list that stopped supply availability of semiconductor products. In particular, this was a huge opportunity for MEMS, which has a different supply chain from the often surprisingly single-source quartz supply chain. Let me explain.
Quartz oscillators are often OEM products, meaning that many companies assemble the products from a variety of suppliers that are often concentrated. A perfect example is the AKM fire, so the supply chain resilience is much lower than at first glance. The 10s of quartz timers company use 2-3 critical component suppliers. This was a huge opportunity for SiTime and MEMS, which relies on Bosch for the MEMS manufacturing and TSMC for the corresponding timing controller.
Because of this time, MEMS likely had one of the greatest adoption periods in the timing industry. Up to this point, MEMS was already growing much faster than the industry, with important design wins in Apple, Tesla, and niche use cases in the Military and Aerospace segments.
But now MEMS won design wins as a second source to quartz and importantly increased the surface area of their customer relationships. AKM’s fire pushed forward the adoption curve rapidly. The problem is that in the near term, this is likely to unwind a bit, as has everything that got pulled forward with COVID; however, I believe that the AKM fire is a critical moment for the further adoption of MEMS.
MEMS is a pretty solid product, and if you notice the gross margins of the companies I listed, SiTime is a league of its own regarding CoGs relative to revenue. Importantly they often price competitive still with quartz, which points to the attractive traits of silicon, and implies an absolute cost advantage over quartz in the long run. There are other advantages I want to discuss, but before I get there, I want to discuss where quartz wins before I outline the bull case for MEMS. I’m trying to keep this balanced.
Quartz is blessed with a low phase noise signal, meaning that random fluctuations don’t mess up the signal over time. The extremely high-end quartz timers offer error rates in the .08ps of error, while the best MEMS can offer is .23ps. Not all products need this level of accuracy, but MEMS has an upward battle to solve the phase noise problem.
There’s a technical core problem here: the MEMS product uses a 40 MHz resonator, and to achieve higher frequencies often needed in networking, they have to multiply their frequency higher. This multiplies noise. A compensating semiconductor here can correct for that, but it adds power.
Another area where quartz often wins out is products above the GHz range. Because of the same noise problem and the 40 MHz resonators, the super high-end frequencies are likely not achievable with reasonable noise and power tradeoff from MEMS. This is another product limitation and one of the blessings of quartz, it can be cut to super high frequencies relative to MEMS and does so with less phase noise.
In an expert call with a research professor, the expert said they do not believe MEMS has a current solution for above 2 GHz. A new resonator or new technology would be needed to achieve that. It’s likely that even if MEMS has a product at 1-2 GHz, the noise, power, and cost might be prohibitive relative to Quartz. Quartz can be cut up to 300 MHz in the standard AT cut and up to 3200 MHz in the SC cut.
This is an incumbency factor, but quartz has been used for 70 years in semiconductor devices. The debugging, errors, and problems in designing quartz into a socket are well understood. That maturity helps for certain designs.
One of the places that MEMS has an absolute advantage over quartz is vibrations. If you remembered the piezoelectric properties of quartz, any vibrations would impact the resonant frequency meaningfully. On the other hand, MEMS is hermetically sealed in silicon and thus a lot stronger at resisting vibration and acceleration issues. MEMS is orders of magnitude better than Quartz in military and space applications. MEMS can withstand 6000+ Gs of acceleration and still tell accurate time. That’s astounding.
This is another area where MEMS has an absolute advantage. Timing oscillators are served on a variety of standard packages quoted in millimeters. Since MEMS is a semiconductor process on the scale of nanometers, they can shrink the package beyond what typical quartz can. An example is the 1.54mm x .84mm package size, meaningfully smaller than the smallest quartz package at 1.6mm x 1mm and often at price parity.
In particular, this is important for smaller IoT devices like headphones, watches, or even smaller devices. Timing is needed for all devices, and MEMS is the only product that offers that small of a footprint.
MEMS is programmable. This might sound confusing, but it comes from the analog device attached to the resonator itself. MEMS have two core frequency resonators, the 32kHz, and 40Mhz resonators, and to serve different frequencies, an advanced compensator mechanism is attached to the resonator. That compensator can be programmed to adjust to the stated frequency needed.
This means that SiTime can buy inventory and then readjust it as needed to the customer’s specs the day before shipping the inventory. This is highly desirable for designers of semiconductor systems, as the changing frequencies and requirements of a complex multi-signal device might require a different frequency. When you design quartz, you are fixed to the 14-week lead time to grow a crystal. If there are any design changes, you cannot adjust that at all.
SiTime has talked about the qualification process being an advantage as well. Because a part is programmable, the designer can customize this unit to whatever need in the design after qualifying a single unit. This is particularly appealing for automotive qualifications. Qualify once and customize later.
The supply chain is oft-cited as an advantage by SiTime management, and the reasoning is twofold. First, most quartz products are partially single-sourced, as shown during the AKM fire supply problems. An advantage is having a true second source supply chain through silicon instead of quartz.
Second, MEMS is a semiconductor process with access to the world's largest and most scalable process. In addition, the lead time and programmability are huge advantages. I mentioned in the programmability section that the ability to swap frequencies mid-design is desirable for semiconductor designers.
This is probably the most contentious piece because, in the real world, MEMS often trades at a premium to quartz or at least a price parity. In my research, I have consistently noted that quartz has a much lower gross margin than MEMS, half of the CoGs in most cases. I believe this implies or confers some kind of price advantage on a like-for-like price basis.
MEMS often can compete in terms of performance, especially in the low end, so they tend to have higher prices with modest and justified premiums over quartz. I believe, however, that if MEMS were to charge price parity, there would be an absolute cost advantage conferred to MEMS because of the semiconductor’s supply chain and scalability of silicon that quartz does not have.
Additionally, something that I think is interesting is that there is likely a road to better prices in the future. MEMS is still an improving technology; for example, SiTime’s process uses 180/90/45nm wafers at TSMC for the circuitry and the MEMs process at Bosch. 300mm wafers for lagging edge, for example, could move their CoGs down meaningfully if greenfield is added. And in addition, the subsequent generations of MEMS resonators have been improving, which could lead to a more performant solution.
Overall I think that MEMS can outcompete in price on the low end and can feature match and price beat at the high end compared to quartz. The improving and scaling semiconductor process is a better cost roadmap than quartz, which seems to be a mature solution without much improvement ahead.
Timing is a niche but multi-billion dollar industry annually, selling 10s of billions of units annually. Timing, like semiconductors, is just one of the niche technology that underpins every aspect of our life. For computers to work together, they must be on the same page, so there’s a timer in almost every digital and analog process of our lives.
Like many other components in 2020, timing also went under a shortage because of the AKM fire. This was just one of the many supply issues faced by the entire semiconductor value chain. But during this critical time, an interesting and new niche technology, MEMS, gained broader adoption.
MEMS resonators and timers were invented in 2006 and comprise a very small part of the total market today. I believe that MEMS, in particular, could grow to be a much larger part of the market as this new technology matures, scales, and faces down broader ecosystem inertia that keeps quartz as the incumbent technology.
To be clear, MEMS has a long road ahead of it - but one that seems pretty clear-cut. I’ll be talking more about the MEMS company (SiTime) in a follow-up post that will include much more detail about the company and the stock itself. I look forward to it next week or the week after.
The next post is about SiTime specifically. I’ll post the investment thesis before a paywall, but the bulk of the writeup, model, and analysis will be behind a paywall. Stay tuned, and please subscribe for the next long piece. I appreciate your support.
Here are a few Tegus calls that I found useful.