Diraq and imec just fabricated an eight qubit array on an ordinary chip factory line

[Candle]

A result published this week in Nature Communications caught my attention more than most quantum papers do, because it is not chasing a flashy qubit count record, it is solving the boring problem that actually determines whether any of this scales. Diraq and imec fabricated and demonstrated an eight qubit linear array on a standard 300 millimetre CMOS silicon foundry line, the same kind of production line that makes ordinary computer chips. That is a meaningfully different achievement from most quantum hardware headlines you see

The reason this matters is manufacturing, not physics in isolation. Most quantum hardware today is built in specialised, low volume labs using techniques that do not remotely resemble how a normal semiconductor chip gets made at scale. Demonstrating silicon spin qubits on a standard industrial CMOS line means the exact same factories that already mass produce classical chips could, in principle, eventually produce quantum processors too, without inventing an entirely separate manufacturing industry from scratch

What impressed me most in the coverage was that the device reportedly held onto solid coherence and control despite being made on infrastructure never originally designed for quantum devices. Scaling quantum dot configurations without sacrificing coherence is exactly the problem that has quietly killed a lot of silicon spin approaches in the past, because standard fabrication tolerances are not built with quantum level precision in mind. Getting eight qubits working cleanly on genuinely standard tooling is a real signal that the tolerances can be met

I want to be honest about scale here though, because eight qubits is nowhere near a useful computer on its own. The significance is entirely about the manufacturing pathway rather than the raw number, similar to how the first transistor on a production line mattered far more for what it implied about scaling than for what it could compute by itself. If this approach holds as they push toward larger arrays, the implication is that silicon spin qubits could piggyback on decades of existing chip manufacturing investment rather than needing an entirely new industrial base built around them

Diraq has apparently framed this as feeding directly into their commercial roadmap, so this was not purely an academic exercise, there is a real company betting on this exact manufacturing path forward. I think that commercial framing is worth noting, because a lot of impressive quantum papers never translate into an actual roadmap a company is willing to publicly commit to

I am curious what the people here who actually understand semiconductor fabrication think about how big a deal the foundry angle really is. Is reusing standard CMOS lines actually the shortcut to scale that the framing implies, or are there hidden problems at larger qubit counts that eight qubits on a line simply cannot reveal yet?

[GameChanger]

I work adjacent to semiconductor fabrication and I think the foundry angle is a bigger deal than most people outside the field appreciate. Building a dedicated quantum fab from scratch is an enormous capital undertaking, so being able to lean on existing 300 millimetre lines that already exist at scale is a real shortcut on the capital side even before you get to the physics

The open question for me is yield rather than coherence specifically. Eight working qubits on a demonstration chip tells you the concept works, but it says very little about what percentage of qubits on a full wafer would actually meet spec at much larger counts, and that yield question is usually where scaling silicon approaches get genuinely painful

[MrRicardo]

Good point on yield, and I would add that standard CMOS tolerances were designed for classical transistors where a small defect is often survivable, whereas quantum coherence can be far less forgiving of tiny variations. So reusing existing lines might get you most of the way there, but the last stretch of precision could still require meaningful custom process tweaks that undercut some of the cost advantage

I do not think that makes this result any less exciting, but I would be cautious about assuming the entire cost advantage of standard fabrication survives all the way to a commercially useful qubit count. Some of it almost certainly does, some of it might quietly evaporate as the requirements tighten

[Vacant Falcon]

Genuine question because I am not a physicist, does coherence at eight qubits actually predict anything meaningful about coherence at say eighty or eight hundred qubits. Or is that a completely separate unknown that this result cannot speak to at all

I ask because a lot of quantum announcements seem to imply a small working demonstration guarantees the bigger version works too. And I have never been clear on whether that assumption is actually physically reasonable or just convenient marketing shorthand

[RedKnight]

It genuinely is a separate unknown in most cases, and that is a fair thing to be sceptical about. Crosstalk between neighbouring qubits, control electronics complexity, and heat dissipation all get meaningfully worse as you add more qubits, and none of those problems are guaranteed to scale linearly just because eight qubits behaved well together

What this result specifically tells you is that the fabrication process itself does not immediately destroy coherence at a small scale, which is a real and necessary first hurdle. It does not by itself prove the harder scaling problems are solved, those genuinely require separate demonstrations at larger qubit counts before anyone can claim victory on scaling specifically

[Daemon82]

I want to add some healthy scepticism about the commercial roadmap framing in the original post, because plenty of quantum companies have announced a roadmap tied to a promising early result that then quietly slipped by years once the harder engineering problems actually showed up. Diraq committing publicly to this path is a positive signal of confidence, but it is not the same thing as proof the path actually works at scale

I would want to see a follow up demonstration at meaningfully more qubits, maybe in the range of fifty to a hundred. Still on standard CMOS tooling, before I updated my confidence in this specific approach very much further than cautious interest

[SortedCougar]

The transistor analogy in the original post is doing a lot of work and I think it holds up better than most quantum industry analogies usually do. The first transistor on a production line mattered because it proved an entire manufacturing paradigm could work, not because that single transistor did anything impressive computationally on its own

If silicon spin qubits genuinely can ride the coattails of decades of existing semiconductor manufacturing investment, that is a structurally different and probably stronger position than architectures that need an entirely bespoke supply chain built from nothing. I think that structural advantage is underrated compared to how much attention raw qubit count records usually get

[Glenn82]

I would gently push back on how much weight to put on any single Nature Communications paper though, because the quantum computing news cycle produces a steady stream of individually impressive sounding results that rarely get properly contextualised against each other by non specialists like most of us in this thread. This could be genuinely important or it could be one interesting data point among dozens published this year alone

I do not say that to dismiss the result, I say it because I think the forum should be a little wary of treating any single paper as a decisive signal about which architecture eventually wins. History in this field has repeatedly rewarded patience over excitement at the first promising demonstration

[Taz18]

On the yield and scaling worries raised earlier in this thread, I think it is worth remembering that the entire modern chip industry solved almost exactly this problem once before with classical transistors, going from unreliable early devices to extremely high yield mass production over a couple of decades. That does not guarantee the same trajectory here, but it is at least a precedent for the specific kind of engineering problem being described

The difference of course is that quantum coherence adds constraints classical transistors never had to deal with at all. So The precedent is encouraging without being any kind of guarantee that the same timeline or the same solutions will apply this time around