Silicon Springs, Microchips, And MEMS

'There's plenty of room at the bottom.' – Richard Feynman, 1959

Silicon balance spring and Microstella balance, Rolex

In 1959, everyone’s favorite bongo-playing hepcat quantum electrodynamics expert, Richard Feynman, gave a lecture titled, “There’s plenty of room at the bottom,” which is often characterized as the lecture that jump started research into nanotechnology. Feynman’s thesis was that one of the most potentially rich fields for research was to look into the possibility of creating and controlling mechanisms built, not with the usual large scale mechanical components, but with molecular and atomic scale components. The paper which published a transcript of the lecture, which you can read here, caused a stir and led to Feynman offering a $1,000 prize to anyone who could create an electric motor, 1/64th of an inch cubed in total volume (not counting wires). The lecture makes for very rich reading like pretty much anything Feynman wrote (I’ve reread his 1985 classic, QED: The Strange Theory Of Light And Matter, several times over the years and like Einstein’s Relativity: The Special And General Theory, it is one of those wonderful science books where a layman can hear about major revolutions in scientific thought, straight from the horse’s mouth, and have a fighting chance of following the arguments).

Much to Feynman’s surprise, the prize was collected in less than a year, by William McLellan (Caltech, 1950) and miniaturization has been producing smaller and smaller mechanisms ever since, leading to the development of what would come to collectively be called the field of MEMS, for Micro Electro Mechanical Systems. (McLellan’s success let to what would eventually become the annual Feynman Challenge, which is still going on today).

A corollary to the paper was Moore’s Law, which was suggested by Intel co-founder Gordon Moore, in 1965; Moore noticed that on average, the number of transistors which you could fit onto a microchip was doubling once every two years, and that this would lead to an exponential increase in transistor density, and therefore microchip power, with no apparent end in sight. Thanks to economic and physical constraints, Moore’s Law no longer really applies, but nonetheless, the density of circuitry in modern microprocessors has put more computing power in the pocket of anyone who owns a mobile phone, than even Moore might have thought possible.

The field of MEMS and the field of microprocessors have one key element in common, and I mean the term “element” literally in this case; the element in question is silicon. Silicon is the 14th element in the periodic table, appearing in the same column as carbon, and it is a “hard, brittle, crystalline solid with a blue-grey metallic luster” as well as being a semiconductor (that is, its conductivity can be made to vary depending on what you do with it, which makes it useful for constructing micro-miniaturized circuits).

Silicon: a “hard, brittle, crystalline solid with a blue-grey metallic luster”

The whole subject is one I started to try and understand in better depth as a result, several days ago, of a bout of insomnia (I get them, on the regular, usually waking up around half past one and finally fall asleep around half past four, and Goldberg, please, play me my variations) in which I was despairing of my own general ignorance of things about which I thought, as a citizen of the modern world, I ought to be better informed. I realized that I had only the most cursory knowledge of how microprocessors work and how they are made, and this despite the fact that my entire professional life, at least insofar as it depends on chips and the Internet, would be impossible without them. It turns out that making microchips is much more expensive, time-consuming, and technically challenging than I’d imagined; chips are made in giant “fabs” or fabricating centers, which consist of huge clean rooms filled with equipment for taking the chips through the different steps of production. I discovered after a few hours, that the processes involved were relevant to understanding why silicon was chosen as a material for balance springs, as well as other escapement components.

Fabbing chips as it turns out takes much, much longer and is much harder than I thought; it takes months, not minutes, hours, or days, to make a chip.

First you need disks of ultra-clean and ultra-pure silicon, which have to be cut from crystals which must be grown in crucibles made of ultra-pure quartz. It is somewhat concerning that 70% to 90% of the high purity quartz needed to make the crucibles comes from Spruce Pine, North Carolina, which is a highly vulnerable source considering how much of the global GDP requires semiconductors (I have no idea exactly how much, but the phrase, “just about all of it” suggests itself). The process involves first etching a base layer of transistors on the chip, which are only three to seven nanometers across – for reference, a single silicon atom is already 0.2 to 0.24 nanometers across and a flu virus is around 80 to 120 nanometers across. The processors are linked with nanometer scale wires embedded in the base layer, and then vertical wires called “vias” are added, which connect the transistors to the next layer up.

This process goes on to the tune of around 80 layers, each of which are made by depositing a new layer of silicon, polishing it flat, covering it with a light sensitive material, exposing the material (the “photoresist”) to UV light shining through a “mask” which carries the pattern of circuitry for the next layer, “developing” the exposed photoresist to remove the areas hit by UV light, removing the residual photoresist (which has to be soft-baked before the the etching stage to remove solvents, and then hard-baked to harden it before the actual etching stage occurs) and then depositing the next layer of nano-scale wiring. The 80 or so reps of the manufacturing process are of course, accompanied by mucho de QC every step of the way, and mucho de cleaning as well, and the design of tools required to make a processor and how you use them are trade secrets like you wouldn’t believe. We don’t give the chips in our phones, laptops, cars, home appliances, and god knows what else much thought but how they are made is as close to magic as anything I’ve ever heard, and the people who know how to make them are, if anyone is, the real powers in this world. God send we don’t do anything to tick them off.

Ingot of pure silicon before being cut into wafers; image, ChipsetC.com

So what does this have to do with balance springs, and other watch components like silicon escape wheels and levers? That’s where the evolution of MEMS comes in.

MEMS has a longer history than you might think. The first transistor was invented by William Shockley at Bell Labs in 1948; it used germanium instead of silicon as a semiconductor, but both silicon and germanium became indispensable in semiconductor research and production. In 1954, C.S. Smith observed that the electrical resistance of silicon changed depending on mechanical stress, and in 1959, the first silicon based pressure sensors were announced. Silicon is an ideal material for micromechanical MEMS sensors, since you can basically etch the mechanical and logic components on one piece of silicon. One of the major breakthroughs in MEMS sensors was the development of accelerometer sensors for air bags, but there are myriad uses – gyroscopic, direction and velocity, and other sensors are staples in mobile devices and many other applications, with one of the key papers being the 1982 “Silicon As A Structural Material” (Petersen). So behind the, to us watch enthusiasts, somewhat sudden and inexplicable appearance of silicon as a structural and mechanical material in watchmaking, is a history going back to the late 1940s and really picking up for the first time in the 1950s, of using it as a micro-mechanical structural material as well.

Everything’s Silicon In The Future!

We hear a lot about the advantages of silicon as a material for balance springs and other watch components and they’re big ones.

A silicon wafer is from one edge to the other, exactly homogenous both chemically and structurally, so a component you make from one part of the wafer is going to be chemically and and structurally identical to a component you make from any other part. Silicon is unaffected by magnetism, and although pure silicon will vary in elasticity as temperature changes, you can coat it with a precisely controlled layer of silicon dioxide which will more or less exactly cancel out temperature related changes. Silicon can be fabricated to nanometer tolerances (we wouldn’t have multi-core CPUs and mobile computing chips if you couldn’t) and one of the great things about it is that mechanical silicon parts can be made with very smooth surfaces.

One of the things I learned whilst poking around the subject, is that there is actually a number that quantifies surface roughness: the Ra number, which is the average of how much the height and depth of peaks and valleys in the surface of a material deviate from the mean. Dig this, hepcats: the Ra value for mirror polished steel is around 0.1 to 0.4 microns, or millionths of a meter; for lapped silicon, it’s less than 0.5 nanometers, or half of one billionth of a meter. Since the purpose of lubricants in the arcane science of tribology is to fill in rough spots, you can see why silicon parts don’t need lubrication; the hills and valleys such as they are, are smaller than the diameter of even some atoms (an atom of gold is 0.3 nanometers, to pick one especially shiny and pretty example). You don’t need to add oil to smooth out something that’s supernaturally smooth already.

This is all great, but it raises the question of why, since the idea of using silicon for mechanical parts has been around since, I would say arguably, at least 1954 (merci, C.S. Smith) it was only starting about 20 years ago that we began to find such parts in watchmaking. One major reason – maybe the major reason – is that to make mechanical parts that will work, not in a MEMS machine, but in a macroscopic machine like a watch movement, the part has to be a lot thicker than a layer in a microprocessor. Individual layers in a microprocessor are sometimes less than a micron thick; a silicon watch balance spring may be more than 50 microns thick.

The relationship between breadth and height is called, in manufacturing as in photography, the aspect ratio. A high aspect ratio in manufacturing means a comparatively narrow width to a comparatively great depth. Making high aspect ratio silicon components for MEMS, or for watchmaking, is hard because you need to make sure that the aspect ratio is constant for the entire depth of the part; it can’t be thicker at the top and narrow to the bottom, or vice versa (unless you want it to be). Making high AR parts was not possible until the invention of etching techniques that let you do it, and the real breakthrough was the development of something you might have already heard of: Deep Reactive Ion Etching, or DRIE.

DRIE uses a hot plasma – a gas so hot that electrons separate from nuclei – which is directed at the surface of the silicon wafer. The resulting ions remove material from the wafer in an almost perfectly vertical direction but they will start to eat into the sidewalls unless you do something about it. The solution is to alternate etching cycles with cycles in which you deposit protective material on the vertical walls, as they get gradually deeper, which prevent the etching ions from eating into the vertical walls and destroying the component.

The Bosch process for deep reactive ion etching, which is the industry standard process for DRIE, was not patented until 1994, and it wasn’t until the process was adopted for watchmaking that silicon components began to appear in watches, with Ulysse Nardin leading the way in terms of putting actual silicon components into watches.

Sidewall of a high aspect ratio silicon component made using the Bosch process; Wikipedia

As with manufacturing microprocessors, what started out as a very high cost, very small production enterprise, has benefited since UN introduced the Freak in 2001, from economies of scale. Today silicon balance springs, as well as silicon levers and escape wheels, are produced for hundreds of thousands of watches every year.

There are indisputable advantages as we have seen, for both customers and watch manufacturers – the cookie-cutter consistency possible with silicon parts, as well as their ability to function without lubricants, and to maintain the same performance characteristics over time, mean longer service intervals, more consistent performance, and better resistance to disturbances from magnetism, positional rate variations, and temperature changes; it’s hard to think of any objection to the widespread adoption of silicon from a practical daily performance standpoint.

What I did get from looking at the evolution of both microchip fabbing, and the evolution of MEMS, is that as with any engineering solution, there are drawbacks as well. Making MEMS silicon components is unquestionably a high tech, industrial process which cannot be duplicated by traditional watchmaking tools and methods, not even close. I’ve written many times that making Nivarox-type balance springs isn’t something a craftsman at their bench can duplicate either – the metallurgy is too complicated, and making such springs at scale requires high tech tools for both production and quality control. (I had never heard of the Greiner Class-O-Matic 3.0 until a couple of days ago, but it’s a pretty sophisticated piece of equipment, used in the industry for measuring the strength of Nivarox alloy balance springs).

Silicon parts are next level, though, in terms of technical challenges, and the difficulty in making them is like the difference between making a Spitfire and an F-35. First of all they wouldn’t exist at all if it weren’t for the supply chain created for the semiconductor industry, which begins with growing ultra high purity silicon crystals in quartz crucibles, the latter of which as you may recall, come from a very small number of obscure (to most of the general public) producers. Secondly, it’s very unlikely they would have been developed without the decades of research and development which led to the industrial production on a large scale of MEMS nanoscale devices for everything from the medical device, to the automotive, aerospace, solar power, and mobile electronics industries. And thirdly, they would be impossible to produce without the invention of techniques like the Bosch process, and other related processes, necessary to produce them. Yes, making Nivarox balance springs is an industrial process but the basic metallurgy was well within the technical capabilities of manufacturers in the 1930s.

It’s true that what man has made, man can repair, at least in theory, but until desktop hobbyist-scale high aspect ration silicon etching becomes possible – if it ever becomes possible – anyone who wants a watch with a broken silicon component serviced, is at the mercy of a highly sophisticated, very elaborate, and probably unreproducible supply chain, on which the entire watchmaking industry is a relatively minor appendage. Unless your idea of DIY or independent watch repair includes in its scope, getting your own quartz crucibles and going from there (bearing in mind that those crucibles are single use only, although you can probably make thousands of balance springs from one rod of pure monocrystalline silicon) you’re going to be depending for the foreseeable future on technologies which are simply not something available to small scale (or hell, even large scale) repair shops, and I wouldn’t want to be around some of the chemicals you need for cleaning and etching silicon parts anyway. Hydrogen fluoride, for instance, is used in the industry for cleaning CVD/Chemical Vapour Deposition chambers, for which I bet it is very effective, considering how exciting fluorine chemistry is in general. And then of course there is the problem of obtaining specs for the parts you want to make at home in your Mattel Easy Etch Fab-O-Matic – every single dimension of any part you want to make has to be specified down to the nanometer, if you want it to work properly, and good luck getting Rolex (or anybody else) to send you the specs for proprietary silicon parts.

The bottom line is that any watch with silicon components is no more something that a watchmaker, as we usually understand the term, is able to repair or replace, than a shot microprocessor is repairable, and the problem gets worse every year. At least in theory, if you have a conventional watch with a broken balance staff, and the balance staff’s made of steel, you can turn a new one on a lathe. If a balance spring needs to be replaced, you can – at least in theory – flatten your own carbon steel wire, wind it to shape, temper it blue, pull the balance out of the watch you want to fix, vibrate the balance and spring until you’re reasonably close to where you want to be in terms of rate, and, after some trouble and effort, you stand at least a chance of getting the watch running again. If you have a Rolex with a Dynapulse escapement – well, here’s what Rolex says about the ceramic Dynapulse balance staff.

“After being machined, the balance staff – made from a white ceramic notable for its extreme strength – is polished to nanometric scale. This polishing is carried out to prevent any fissures or damage in the event of impact, while creating a perfectly smooth surface finish. Shaping the staff by means of a femtosecond laser also allows the creation of optimal geometry for the pivots – the points at each end of the staff – thus enhancing their shock resistance.”

None of this sounds especially tractable to the old timey watchmaker at their bench but when you get to shaping the staff by means of a femtosecond laser the sense of utter inadequacy in the face of scifi-level industrial processes really started to sting. God knows how they attach the balance to the staff – transdimensional nanoprobes manipulating Planck-scale black holes, for all I know.

Watchmaking has relied a lot, historically, on a heritage and history message and at least from a design standpoint, that’s still a message that makes sense, but in terms of connecting to the history of actual watchmaking – well, those arts and crafts are actually things that in many cases, the industry is abandoning as fast as it can. And you thought CNC machines were bad.

Coda: A millennial makes a modest proposal. That the world is in a pretty fix in a lot of different ways these days should be an uncontroversial statement but of course, there is plenty of room to argue over the details. If you’re curious about what the problems and solutions look like to one twenty-something, I’d like to recommend checking out a literary exercise produced by my older son, Zach, who has a lot of interests/worries which are common to many of us (the role of AI in shaping the future, the deranged state of modern political discourse) and some which are a little bit less widespread (the history of the Ancient Near East, Sumerian). He has a Substack of his own and not long ago, he wrote what I think is an interesting exercise in literary revisionism, taking Sun Tzu’s The Art Of War as a starting point. His basic thesis is that while wise Sun Tzu’s writing is often looked at for pointers on the conduct of civil governance, that maybe using a manual on military tactics and strategy might not be such a hot idea if your aim is a stable and peaceful society. His “manifesto” as he calls it, somewhat tongue in cheek, is part parody of Sun Tzu, and partly critique; you can find The Art of Governance: A Counterpart to Sun Tzu’s The Art of War, Focusing on Matters of Peace, right here.

Coda Part II. If you’re interesting in going really in-depth on the history of MEMS, check out History of MEMS, Primary Knowledge Participant Guide, which was essential for this story. For a general history of silicon in watchmaking, in the context of balance springs, I recommend as a great starting point, Hairspring Evolution And Materials, Part II, at Watches By SJX. If you’re interested in a very solid, well produced and informative introduction to the intricacies of microprocessor fabrication, the video How Are Microchips Made? CPU Manufacturing Process Steps is a must-see. And for a closer look at the production of silicon crystals and wafers, see History of Silicon Wafers, at ChipsetC.com.

A note to subscribers: Split Seconds is reader supported via paid subscriptions, but as always, some wider interest stories are free to read, including this one, which reflects a number of considerations not the least of which is my ambivalence about paywalls. Thanks for your support, and see you behind the paywall next time, amigos.