One of the challenges of making useful quantum computers is the regular computers are already so goddamn powerful. However you look at it, silicon electronics has had a run unlike anything else. Trilions of dollars were invested to develop trillions of dollars’ worth of capabilities which bring customers trillions of dollars in value.

Now say you want to build a quantum computer. Unless you have trillions of dollars of R&D funding, you probably want to piggyback of some of that CMOS tech, right? That definitely sounds like the way to go. But how far can you push it?

(image source: a news website which did not quote its source either)

At the APS March Meeting, it seemed like everyone was trying to convince you that their quantum computer is CMOS compatible. But is it?

Silicon spin qubits

The main contender for CMOS compatibility are, unsurprisingly, semiconductor spin qubits. Following seminal work in A CMOS silicon spin qubit | Nature Communications, this has picked up popularity and now plenty of players are jumping on that bandwagon. So are they “fully CMOS compatible”? It feels the answer is “generally yes”, though a pedant could make a few caveats:

  • There are a number of techniques used to fabricate and control quantum dots, and while some of them are CMOS compatible, it remains to be seen whether the highest-fidelity operations can happen in CMOS devices.
  • Intel’s silicon spin qubits are “not fully CMOS”, as they actually pattern cobalt micromagnets on top of the chip post-fab
  • For high performance, it is insufficient to use “standard silicon” - one needs isotropically enriched silicon instead

Photonic qubits

In response to Intel’s talk, Matthew House from PsiQuantum gave a talk proclaiming something to the extent of: “we reject the notion that silicon qubits are the only CMOS-compatible platform - the same is true for photonic qubits!”. This sounds reasonable - silicon waveguides sure do sound CMOS compatible! However, a pendant might poke a few holes in that story:

  • While silicon photonics is fairly standard, a lot of photonic qubits want to use silicon nitride (SiN). While SiN is a CMOS-compatible material, SiN waveguide litography is not a well-established CMOS process, especially at low temperatures
  • Lowest-loss photonic interconnects require edge coupling, and are typically done by hand
  • PsiQuantum decided their modulator of choice are epitaxially grown BTO thin films. Not exactly your typical multiproject wafer…

Just to emphasise, there is nothing wrong with this approach. Realistically, it is unlikely that quantum computers require the exact same processes and materials as classical computers. So while one may start with “I’m fully CMOS compatible”, the likely end point is “I’m CMOS compatible, as long a we tweak CMOS a bit”.

Trapped-ion qubits

The third candidate for a CMOS-compatible quantum computer are trapped ions, though they rarely advertise as such. Still, it it well known that you can order an ion trap chip from a CMOS foundry and trap ions on top of it (see for example [1406.3643] Ion traps fabricated in a CMOS foundry). CMOS techniques are also compatible with many more advanced trap designs, such as chips with integrated detectors or voltage sources. So are ion traps “fully CMOS compatible”? As before, I’d say “yes, except some caveats”, for example:

  • Scalable optical delivery requires SiN and possibly other waveguide materials - which are CMOS compatible, but with caveats above.
  • Ion trap operation can be disturbed by surface noise. Can surfaces coming out of CMOS processes have noise levels comparable with state of the art? Probably, but we don’t know.

Superconducting qubits

Now let’s look at a classic “non-CMOS” platform: superconducting qubits. State-of-the-art transmons require all kinds of weird materials and fabrication processes that were optimised in cleanrooms across the world to minimise decoherence. So, are superconducting qubits “not CMOS comparible”? Not quite!

Neutral atom qubits

Another classic non-CMOS platforms are neutral atoms in tweezer arrays. Here, you trap and manipulate hundreds of atoms in free space in vacuum by shining light through big lenses and spatial light modulators or acousto-optic deflectors. Does not sound very much like a classical computer, does it? But is it really “not CMOS compatible”?

  • Another view of the same platform is that the atoms are only a nonlinear element in an otherwise photonic circuit: laser light comes in, gets split between multiple individually modulated channels, is focussed onto the nonlinear element, and converted into an outgoing photon which is detected and processed. And while light focussing and collection may be a bit “non-CMOS”, other parts may well be.
  • I think this is the kind of thing that Dirk Englund’s group has been getting at lately in works like this one: [2210.03100] Scalable photonic integrated circuits for programmable control of atomic systems. Just like ion traps are a “CMOS chip plus ions above”, neutral atoms can, one day, be a “CMOS chip plus atoms above”.

The same arguments can be made about colour centres. You probably cannot fabricate them in a CMOS chip, but how about simply attaching on top of your CMOS PIC? See [1911.05265] Large-scale integration of near-indistinguishable artificial atoms in hybrid photonic circuits.

Summary

So the next time someone asks you to trust them, because their quantum computing platform is CMOS compatible, remember there is more subtlety to this claim than meets the eye!