Stabilizer 5

[Emerson Mata]

Fordecades this has been the core absurdity of luxury typing: you can buy a $4,000 hyper-engineered super-premium board with exotic materials, rare hybrid switches, hidden fasteners, leaf springs, and custom gaskets interfacing every surface. Yet among its most critical components can be found a fundamentally flawed mechanism that has seen little refinement since its inception.

Below is the architecture of a traditional Cherry-type stabilizer. A keycap is connected to two sliders via a press fit onto their cruciform mounting points. Those sliders move up and down with the key travel, and a spring-loaded switch (also pressed into the keycap) is situated between the two stabilizing ends. A rotating wire clipped into the stabilizer housings has two extension arms that insert into the two sliders, pulling them up and down together in tandem and thus creating the anti-seesaw effect.

There are two primary sources of unwanted sounds. I'll formalize some keyboard community terminology here by calling the first rattle, which refers to the sound of the hard wire linkage striking against the hard plastic surfaces in the interior of the slider. This is by far the most prominent and displeasing source of stabilizer noise.

There is one additional complicating factor, which is that long keys like spacebars frequently vary between manufacturers in their stem spacing length. Any widely applicable stabilizer design must therefore accommodate a remarkable degree of dimensional variability.

It is, in a sense, not a fundamentally untrue observation that fits are an important part of the issue. But were it down to gap size alone, changing a few numbers on a spec sheet would have done the trick, and the problem would have been solved long ago. The fact that it hasn't, in fact, been solved long ago is therefore instructive.

Indeed, if we are to design a perfect stabilizer, we must first confront a contradiction in our requirements: we need the mechanism to be at once loose and tight. It must be "tight" in the sense that there must be no backlash or movement between parts that could create unwanted sounds, but also in the sense that it must accurately and with minimal wobble perform the anti-seesaw action that is the foundational premise of a key stabilizer. These objectives typically require small gaps between mating parts. But the stabilizer mechanism must at the same time also be "loose" in the sense that the key must move up and down very freely to avoid any chance of binding, which is when the key ceases to move due to accumulated forces within the system (most commonly causing the keycap to fail to pop freely back up after being depressed). To achieve this objective, large gaps between mating parts are required. Thus the contradiction.

It may not be obvious on casual inspection, but a key stabilizer is essentially a sliding linear bearing: that is, a mechanism to provide unrestricted translational motion along an axis, while restricting motion in other axes. One common example of a linear bearing would be a rectangular carriage that moves along the length of two cylindrical rods.

Linear bearings are well characterized in the world of mechanical engineering and are subject to two important design considerations known as bearing ratio and binding ratio. These consider the geometric relationship of variables like the distance of the bearing load to the length of the bearing. As with all engineering rules of thumb, these ratios give a practical handhold on much more complex points of underlying physics. Rather than delve here into force vectors and friction calculations, which I don't understand fully myself anyway, suffice it to say that when the rules are violated, bearings become very sensitive to friction and manufacturing variation, and the accumulation of forces inevitably leads to binding and other performance problems (See Design of Machinery for a more in-depth discussion.)

It's enough to know that the geometric constraints of a traditional stabilized key switch are a particularly bad scenario for the rule-of-thumb ratios: a very short travel distance, with both a long "moment arm" (in this case, the keycap) and large diameter relative to the bearing length. These are all the exact opposite of what we want for good linear bearing performance, and thus what we can expect from the theory is exactly what we find: most solutions are extremely sensitive to manufacturing variation and prone to binding.

Of course, I didn't really understand any of this when I embarked on an attempt to solve the stabilizer problem. I have no formal training in mechanical engineering, but I tend to believe in the power of empirical tinkering, and have often been surprised by how far that has been able to get me in many domains of life. It was some combination, then, of ignorance and yolo audacity that allowed me to think it could somehow be an easy problem to solve.

It should be noted that, by the time I began this project, the Seneca keyboard for which these stabilizers were intended had already surmounted a number of hard engineering hurdles, not least of which was the creation of my own capacitive PCB platform and firmware. The overall keyboard was, as I perceived it then, already 90% done. The stabilizers were merely "the last little 10%" to get the whole product to be perfect.

Oddly, my first foray into this problem around 2019 began with the realization that a stabilizer is in fact a linear bearing (without knowing any of the corresponding design rules of thumb). At a manufacturing trade show, I had seen some interesting "self-lubricating" plastic bushings for linear bearing rods, made solid out of a special injection-molded polymer, and so it occurred to me to give those a try. These seemed like a good option because the tiny work envelope of keyboard stabilizers requires much smaller parts than are typically manufactured as ball or roller bearings, which would otherwise be the natural bearing type to use for something like this. I figured if we were just to create a cruciform mounting point to insert some low-friction cylindrical rods into the keycap stems, make those rods with precision, and then insert those rods into parallel bushings inserted into a switch plate, all would be well; problem solved. I knew from the spec sheets that these bushings required a particular (tight) tolerance range for a free-running clearance fit, and in attempting to make rods to fit into them (and attach to the keycap stems) I quickly learned the first lesson of how difficult the road ahead would be.

Tiny parts are extremely expensive to prototype with accuracy and good surface finishes. The only machine shop I could find that would make the prototype parts for me was one that specialized in manufacturing for Swiss watch makers, at the cost of some thousands of dollars for just a few test samples. This is a fact I would go on learning as I would continue to refine my experimental designs. Even 3D printed parts made to the right tolerance and surfaces I needed to get good information would be insanely expensive to produce, and thus the cost of iterating much higher than if the parts didn't have such minuscule mission-critical features. The below screenshot shows orders of parts that that, together, were all small enough to fit in the center of the palm of my hand, and I'd be embarrassed to admit how many such orders I've placed over the past few years.

I did also eventually get a Formlabs SLA printer for in-house prototyping, but the surface finish and fine details were only good for very rough initial concept exploration, requiring the outlandishly expensive Microfine printing process above for more detailed studies. SLA resins are all very susceptible to abrasion and so only a few switch actuations of the parts sliding against each other would change their dimensions and roughen their surfaces. Even though the finishes right out of the printer were as good as almost any 3D printing technology can yield, it was still much rougher than actual injection molded parts. This would greatly increase friction and exaggerate binding issues, so I had to brush XTC-3D carefully onto the surfaces to get something even approaching a production surface and thus a more accurate sense of binding. Maddeningly, this would in turn throw off the part dimensions and tighten up the gaps, itself an independent cause of unrepresentative binding. Test molding the parts in thermoplastics was an option, of course, but often at the cost of more than $10,000 per part, and quick-turn companies like Protolabs are extremely limited in their abilities to make parts with side actions and high gloss surface finishes, so most of what I wanted to try could only be done with very slow test molding through traditional factories (a process generally with a multi-month turnaround).

What was I was experiencing without realizing it were the effects of an awful bearing and binding ratio along with kinematic over-constraint (on which more later). All I knew at the time, though, was that it wasn't an auspicious beginning.

I floundered around for months, talking to various mechanical engineers and examining if there might be ways to go back to the basic existing stabilizer paradigm and perform some modest tweaks to mitigate sound issues.

One later concept I came upon, which wasn't entirely off base, involved the use of a Scotch yoke design otherwise similar to existing stabilizers. By inserting the wire from the side and allowing it to rotate in a slot within the slider, it would be theoretically possible to have a smaller clearance gap between slider and wire than in front-insertion traditional stabilizers.

This got acceptable results, but it required very large gaps between slider and housing to get free movement without binding. Rattle was slightly improved, but ticking was, if anything, worse than on traditional stabilizers. And, every time I would tighten up the interfaces to improve acoustics, the system would bind up.

3a8082e126