For decades, the dream of building a computer that thinks the way a brain does has run into a stubborn physical wall: heat. Brains are astonishingly frugal, firing trillions of signals on roughly the power of a dim lightbulb, while the silicon we use to imitate them tends to run hot and hungry. A team at the University of Hong Kong has now found an unexpected place to chase that efficiency — not in the warmth of a data center, but in the deep cold, a hair above absolute zero, where almost all conventional electronics stop behaving the way their designers intended.
What Happened
Researchers from the Department of Electrical and Computer Engineering in the Faculty of Engineering at the University of Hong Kong (HKU), together with the Centre for Advanced Semiconductors and Integrated Circuits (CASIC), have built a programmable neuromorphic hardware platform that operates near absolute zero. The work, led by Professor Yuhao Zhang and PhD student Xin Yang, was published in Nature Communications under the title “Cryogenic neuromorphic circuits using gate-controlled negative differential resistance in silicon carbide.” HKU announced the result in early June 2026, and it quickly drew attention across the quantum and semiconductor press.
The heart of the discovery is a way to coax a familiar industrial component — the silicon carbide (SiC) MOSFET, a transistor already mass-produced for electric vehicles and power grids — into behaving like a biological neuron. The team demonstrated that a single transistor can reproduce the energy-efficient “spiking” behavior of a neuron at temperatures as low as 10 millikelvin, colder than the deepest reaches of interstellar space. They achieved this by generating and controlling a property called negative differential resistance, or NDR, inside the device.
What makes the approach elegant is where the effect comes from. When the SiC transistors are cooled below about 2 kelvin, they exhibit a pronounced “S-shape” NDR driven by a mechanism the researchers call electron-donor impact ionization. Crucially, this behavior is intrinsic to the material’s atomic structure rather than something engineered on top of it, which the team says makes it stable and repeatable from one manufacturing batch to the next. The spiking neurons can also be cascaded together into larger networks, hinting at circuits that process information locally rather than shuttling every signal back to a warm controller far away.
Why It Matters
To appreciate why a cold-loving neuron chip is more than a laboratory curiosity, it helps to look at the bottleneck quantum computers are running into. Qubits, the fragile building blocks of a quantum machine, must be kept at millikelvin temperatures inside elaborate dilution refrigerators. But the electronics that read and control those qubits are conventional silicon chips that generate heat and draw substantial power. To keep that heat away from the qubits, engineers place the control hardware far from the quantum processor and connect the two with dense bundles of wiring.
.jpg){kind=link}
That separation has quietly become one of the central obstacles to scaling quantum computers up from hundreds of qubits to the millions that genuinely useful machines may require. Every additional qubit needs its own control lines, and the wiring grows into a tangle that is difficult to cool, expensive to build, and limiting to performance. Professor Zhang frames the HKU platform as a way to break that pattern, describing circuits that can sit right next to the quantum processor and run, in his words, “thousands of times more energy-efficient than conventional electronics.” By generating so little heat, the neuromorphic chips could be integrated alongside the qubits instead of being exiled to a warmer stage of the refrigerator.
There is also a manufacturing argument that gives the work unusual practical weight. Silicon carbide is not an exotic material that needs a new industry to support it; it is already produced at scale on 300-millimeter wafers in existing foundries. That means, at least in principle, the path from a published result to a fabricated chip does not require inventing an entirely new supply chain — a recurring stumbling block for promising laboratory devices that never reach production.
Reaction
The announcement landed at an interesting moment for neuromorphic computing, a field that has spent years promising brain-like efficiency but has mostly delivered it at room temperature. Outlets covering the quantum and semiconductor industries highlighted the result as a “world-first” cryogenic neuromorphic platform, and specialist publications framed it less as a finished product than as a proof that an important idea works at all: that a standard transistor, cooled hard enough, can be turned into a spiking neuron without bolting on extra materials.
Part of the enthusiasm comes from the fact that the team leaned on an industrial workhorse rather than a custom-grown novelty. Xin Yang, the PhD student who co-led the work, emphasized the pragmatism of building on silicon carbide, noting that because the material is already used globally, existing foundries could fabricate the cryogenic chips. For a research community that has watched many beautiful devices die on the way to the fab, that grounding in real manufacturing is part of what made the result resonate.
It is worth keeping the excitement measured. The work demonstrates single neurons and small cascaded circuits, not a complete control system wired into a working quantum computer. The leap from a handful of spiking transistors to a dense, reliable layer of cryogenic intelligence sitting beside thousands of qubits is a long one, and the paper is best read as an opening rather than a conclusion.
What's Next
The most immediate target is quantum control. If neuromorphic circuits can perform real-time processing right at the cold heart of a quantum machine, they could help with quantum error correction — the relentless, computation-heavy task of catching and fixing the mistakes qubits constantly make — and with the fast feedback loops needed to keep qubits coherent. Doing that processing locally, instead of sending raw signals up and down long wires, is exactly the kind of architecture that might let quantum systems grow without drowning in their own cabling.
The second frontier is further away, literally. Electronics destined for the Moon’s surface or the outer solar system must survive brutal cold, and circuits that are not merely tolerant of low temperatures but actually designed to thrive in them are rare and valuable. The HKU team points to deep-space exploration as a natural home for these rugged components, where the same property that makes them awkward on a warm desktop — their love of the cold — becomes a genuine advantage.
Between those two horizons lies the unglamorous work that determines whether a breakthrough becomes a technology: scaling the circuits up, integrating them with real qubit hardware, and proving that the stability seen in the lab holds across full wafers and long operating runs. Those are the questions the next round of research will have to answer.
Closing Thoughts
There is a quiet lesson in this result about where progress sometimes comes from. The instinct in computing has long been to push toward more heat tolerance, faster clocks, denser packing — to make machines that perform in spite of physics. The HKU work goes the other way, finding a regime so extreme that a humble, mass-produced transistor starts to behave like something alive, and then asking what that behavior is good for. Efficiency, it turns out, can be a property you discover rather than one you force.
Whether cryogenic neuromorphic chips end up controlling the quantum computers of the 2030s or riding aboard a probe to the outer planets, the deeper point is that the boundary between biological and artificial computation keeps thinning in unexpected ways. A neuron that fires at ten thousandths of a degree above absolute zero is a strange and beautiful thing to have built — and a reminder that the most interesting machines often emerge where we stop fighting nature and start listening to it.
한글 요약
홍콩대학교(HKU) 공학부 연구진이 절대영도에 가까운 극저온에서 작동하는 ‘뇌를 닮은’ 뉴로모픽 반도체 플랫폼을 개발해 Nature Communications에 발표했다. 위하오 장(Yuhao Zhang) 교수와 박사과정생 신 양(Xin Yang)이 이끈 연구진은 전기차와 전력망에 이미 대량 생산되는 실리콘 카바이드(SiC) 트랜지스터에서 ‘음의 미분 저항(NDR)’ 현상을 제어하는 방법을 찾아냈다. 이를 통해 단일 트랜지스터 하나가 10밀리켈빈이라는 극한의 저온에서 생체 뉴런처럼 에너지 효율적인 ‘스파이크’ 신호를 흉내 낼 수 있음을 처음으로 입증했다.
이 기술이 주목받는 이유는 양자컴퓨터의 핵심 난제와 맞닿아 있기 때문이다. 양자컴퓨터의 큐비트는 극저온에서 유지돼야 하지만, 이를 제어하는 기존 실리콘 전자장치는 열을 많이 내뿜어 큐비트에서 멀리 떨어뜨려 놓아야 하고, 그 결과 복잡한 배선이 확장의 발목을 잡아 왔다. HKU 플랫폼은 기존 전자장치보다 수천 배 에너지 효율이 높아 큐비트 바로 옆에 둘 수 있어, 양자 오류 정정과 실시간 제어의 병목을 풀 잠재력이 있다. 게다가 SiC는 300mm 웨이퍼 기반 기존 파운드리에서 바로 양산할 수 있다는 점도 강점으로 꼽힌다.
다만 이번 연구는 단일 뉴런과 소규모 회로를 시연한 단계로, 실제 양자컴퓨터에 통합된 완성형 제어 시스템은 아니라는 점에서 신중하게 볼 필요가 있다. 연구진은 양자 제어를 넘어 달 표면이나 외태양계처럼 극한의 추위를 견뎌야 하는 심우주 탐사 전자장치로의 활용도 내다보고 있다. 추위를 약점이 아니라 강점으로 바꾼 이 접근법이 어디까지 확장될지가 다음 과제다.
참고 / 출처: HKU 보도자료, Nature Communications 논문, The Quantum Insider, ScienceDaily