Brain Inspired Chips Greener Data Centers Explained: Post-CMOS Hardware Neurons Take on the Memory Wall (World Tech)

Image: Wikimedia Commons / Carl Lender, CC BY 2.0 — server racks in a data centre (illustrative; the review's brain-inspired hardware aims to cut exactly this kind of energy load).

TL;DR — brain inspired chips greener data centers explained: researchers at the University of Ottawa, Columbia University and Hanyang University have mapped how "post-CMOS" hardware neurons — chips that compute the way brains do, with memory and processing in the same place — could break the memory wall that makes today's AI data centres so power-hungry. It is a research roadmap, not a product: no chip from this review is on sale, and no data-centre energy saving has been measured yet.

📌 KEY HIGHLIGHTS

• The work: a systematic review, "Emerging Post-CMOS Hardware Neurons for Brain-Inspired Computing", published in Advanced Functional Materials in 2026 (DOI: 10.1002/adfm.77874).
• The team: lead author Dr Kannan Udaya Mohanan, in Professor Chang-Hyun Kim's group at the University of Ottawa, with Columbia University and Hanyang University.
• The problem: the "memory wall" — conventional chips waste time and energy shuttling data between separate memory and processors.
• The idea: hardware neurons that add up signals and fire like brain cells, so memory and computation live together and the chip only works when something changes.
• The honest limit: this is a roadmap of candidate materials and architectures, not a finished chip — large-scale manufacturing and real-world savings are still unproven.

What happened?

On 21 October 2026, the University of Ottawa's research news carried a simple pitch: if AI data centres are eating ever more electricity, maybe the answer is to stop building computers like filing cabinets and start building them like brains. Behind the pitch is a systematic review, not a single experiment. Professor Chang-Hyun Kim's team — led by postdoctoral fellow Dr Kannan Udaya Mohanan, with collaborators at Columbia University and Hanyang University — spent years of published work in electronic materials, device physics and circuit integration and distilled it into one document: a map of the most promising "post-CMOS" materials and device architectures that can act as artificial neurons, and of how those devices could be assembled into scalable, brain-like computing systems.

The review was published in the journal Advanced Functional Materials, and the story was picked up again on 23 October by independent coverage noting the same core claim: that treating the brain as the model for the whole chip architecture, rather than for one component, is what sets this roadmap apart from the many one-off neuromorphic prototypes of recent years. That distinction matters, because it is also the review's biggest caveat — a roadmap points somewhere; it does not arrive there.

Brain inspired chips for greener data centers explained: how a hardware neuron works

A biological neuron is, at heart, a tiny adding machine with a tripwire. Signals arrive from other neurons, the neuron adds them up, and if the total crosses a threshold, it fires — sending a sharp electrical spike onward. A hardware neuron copies that shape in electronics. Incoming signals are summed, a non-linear "activation" decides whether the total matters, and the device emits a spike only when it fires. Synapses — the connections between neurons — are arranged in grids called crossbar arrays, where each crossing point holds a weight that strengthens or weakens with use, the way a memory forms.

That is the opposite of how your laptop works. A conventional processor and its memory live apart: the processor fetches data, computes, and writes the result back, millions of times a second, across wires that cost time and energy on every trip. Engineers call this the memory wall, and as AI models have grown, the wall has grown with them — moving data, not crunching numbers, is now a large share of what a data centre's power bill pays for. A brain-inspired chip attacks the wall directly by refusing to move the data at all: memory and computation sit in the same device, and because spikes are sparse and event-driven, large parts of the chip sit idle — using almost nothing — until something actually changes.

The comparison the researchers keep returning to is the brain itself, which runs a whole thinking machine on a power budget closer to a household light bulb than to a server rack. Nobody is claiming a chip built next year will match that. The claim is narrower and more interesting: that the architecture — co-located memory, spike-based signalling, adaptive weights — is the right direction, and that the review identifies which emerging materials and device designs are the most credible ways to get there.

Image: Wikimedia Commons / Birdmanoftech, CC BY-SA 4.0 — diagram of neural processor units in a mesh array (illustrative of neuromorphic architecture; this is BrainChip's AKD processor diagram, not a figure from the Ottawa review).

Traditional chips vs brain-inspired hardware neurons

Feature	Traditional processor (CMOS)	Brain-inspired hardware neuron (post-CMOS)
Where memory lives	Separate from the processor; data is fetched and returned on every operation	In the device itself, as synaptic weights in a crossbar array
How it computes	Clock-driven: every unit works on a fixed schedule, whether needed or not	Event-driven: units fire spikes only when their input crosses a threshold
Energy behaviour	Constant power draw, plus the cost of moving data back and forth	Mostly idle until triggered — energy spent only where something changes
Strength	Mature manufacturing, vast software ecosystem, predictable performance	Potentially far lower energy per task; adaptive, learns from patterns
Limit today	The memory wall: data movement wastes time and power	No proven large-scale manufacturing; software and benchmarks still catching up

"Post-CMOS" simply means "beyond the transistor technology every chip today is built on". CMOS — complementary metal-oxide-semiconductor — is the standard way of making chips, and it has carried computing for decades. The review's candidates go beyond it: emerging materials and device structures that can behave like neurons and synapses directly, rather than simulating them with millions of conventional transistors. The same family of materials, the authors note, may also be lower-cost, highly scalable, and compatible with ultrathin, lightweight and flexible shapes — which is why the roadmap extends past data centres into wearables, personalised health devices and more natural human-machine interfaces.

Why it matters for India readers

India is adding data-centre capacity fast, and every new facility lands on the same two constraints this review is about: electricity and cooling. Power is already one of the largest operating costs for Indian data centres, and AI workloads make it worse — training and running large models means more servers, more heat and more grid demand. A hardware direction that cuts energy per task would matter in India earlier and more sharply than in countries where power is cheap and plentiful, because here the saving shows up directly in operating costs and in whether a facility can even get a grid connection.

The second India angle is at the other end of the scale: small, low-power, always-on devices. India is one of the world's largest markets for wearables and affordable health tech, and those products live or die on battery life. Brain-inspired hardware that sips power while listening, sensing and adapting — instead of waking a power-hungry processor for every task — is exactly the kind of chip a future generation of Indian wearables and health monitors would be built around. The review itself names no Indian project, and this post claims none: the point is that the constraints it addresses — power, heat, cost — are India's constraints too.

Not a product, and not living brain cells

Two misunderstandings are worth clearing up, because both are already circulating around "brain chips". First, nothing in this review is a product you can buy, and no energy saving has been measured in a real data centre. The authors are explicit that scalable circuits are the necessary next step, and the independent coverage agrees: moving from laboratory devices to large-scale production, and proving compatibility with existing semiconductor manufacturing, both remain to be demonstrated. Treat any headline promising a percentage energy saving from this review as hype — the review promises a direction, not a number.

Second, these are silicon-and-materials chips, not living tissue. A different, much smaller field builds computers from lab-grown neurons on electrodes — the "biological computers" you may have seen playing simple video games. That is real research, but it is not this review. Post-CMOS hardware neurons are electronic devices that imitate how neurons behave; no cells, no life support, no biology involved. Confusing the two makes the story stranger than it is — and hides the genuinely interesting engineering question, which is whether electronics can be made to compute the way brains do.

Should you care?

✅ Yes — if you follow AI infrastructure, semiconductors or energy: this is the clearest recent map of the post-CMOS landscape, from a team spanning Canada, the US and South Korea, and it frames the data-centre energy problem as an architecture problem, not just a "build more power plants" problem. ⏳ Watch the manufacturing step: the roadmap only matters if these devices can be made at scale in existing chip fabs — that is the test to track over the next few years. 🔍 Honest caveat: this is a review, not an experiment — there are no new benchmark results here, and the energy gains are projections, not measurements.

FAQs

What are brain-inspired chips?
Chips that copy the brain's way of computing: artificial neurons add up incoming signals and fire spikes only when a threshold is crossed, and memory lives in the connections (synapses) rather than in a separate memory chip. They are also called neuromorphic chips, and the goal is to compute with far less energy than conventional processors.

What does post-CMOS mean?
CMOS is the standard transistor technology used to make virtually all computer chips today. Post-CMOS means materials and device designs that go beyond conventional CMOS transistors — devices that can act as neurons and synapses directly, instead of simulating them with millions of ordinary transistors.

What is the memory wall in computing?
The bottleneck caused by keeping a processor and its memory in separate places. Every computation forces data to travel back and forth between them, and that movement costs time and energy — increasingly more than the computation itself. Brain-inspired chips attack the wall by putting memory and processing in the same device.

Are brain-inspired chips the same as computers made from living brain cells?
No. This review is about electronic devices made from emerging semiconductor materials that imitate neuron behaviour. Computers built from lab-grown neurons on electrodes are a separate, much smaller research field — interesting, but not what the Ottawa review describes.

When will brain-inspired chips reach data centres?
There is no date. The review is a roadmap, not a product announcement: the authors say scalable circuits are the necessary next step, and large-scale manufacturing plus compatibility with existing chip fabrication still have to be demonstrated. Expect edge devices and specialised hardware first, and data-centre impact only after the manufacturing question is answered.

Takeaway: The Ottawa review does not give us a new chip — it gives us a map. Its message is that the data-centre energy crisis is, at root, an architecture problem: stop shuttling data between memory and processor, and start computing where the data lives, the way brains already do. The map is credible; the journey to a factory is still ahead.

Follow World Tech on TechByCountry for tomorrow's update.

📸 Get tech in the last 24 hours on Instagram — follow @_techinlast24hrs for daily posts (9 AM–9 PM) and reels at 12:30 PM & 8:30 PM.

Sources

TechXplore — Building brain-inspired semiconductor chips for smarter, greener data centers (21 Oct 2026)
ScienMag — Research report: Building brain-inspired semiconductor chips for smarter, greener data centres (21 Oct 2026)
alextech.ai — Brain-inspired chips could cut AI data center power use (23 Oct 2026)
BrightSurf — Building brain-inspired semiconductor chips for smarter, greener data centres (21 Oct 2026)
Kannan Udaya Mohanan et al., "Emerging Post-CMOS Hardware Neurons for Brain-Inspired Computing: Devices, Circuits, and System Integration", Advanced Functional Materials (2026), DOI: 10.1002/adfm.77874

Related TechByCountry posts: Silicon slow light modulator 200 Gbps explained, 7 eV bandgap semiconductor explained, Microcavity polariton superfluid Bose insulator explained.