Silicon Minds

Neuromorphic computing and India’s opportunity to leapfrog to the physical AI era

Every time a large language model answers a query, it consumes roughly the same energy as charging a smartphone. Multiply that across billions of daily requests, and you have an industry sprinting toward a wall: AI’s insatiable compute demands are already straining power grids, inflating data centre costs, and making the economics of inference—never mind training—structurally unsustainable.

The root cause is architectural. Modern AI runs on hardware designed in the 1940s. The von Neumann architecture physically separates the processor from memory, which imposes a quadratic energy and latency penalty on parallel AI workloads: double the task, quadruple the cost. Decades of Moore’s Law papered over this inefficiency, but silicon scaling is now approaching its physical limits. The next leap in AI compute will require not merely faster chips, but fundamentally different ones.

The Brain-Inspired Alternative

Neuromorphic computing takes its design cues from biological neural networks. Rather than shuttling data back and forth between a processor and separate memory, it co-locates computation and storage inside synthetic neurons and synapses—as the brain itself does. The result is computation that is massively parallel, event-driven, and orders of magnitude more energy-efficient than conventional silicon.

The commercial case is sharpening. Always-on edge devices—satellites, drones, medical implants, industrial sensors—cannot connect to a megawatt data centre and operate with power budgets measured in milliwatts. Neuromorphic architectures are uniquely suited to this constraint. At Speciale Invest, our investment thesis has long been anchored in backing efforts that dismantle the physical limits of mainstream technology. Our investment in Morphing Machines, which targets the von Neumann bottleneck directly, reflected this conviction early. Neuromorphic computing is the next frontier in that arc.

$213M MARKET 2025

→

$1.6B PROJECTED 2034

The neuromorphic computing market grows 7× over the decade—but the more significant figure is what the technology unlocks: edge AI, sovereign compute, and always-on inference at scale. Source: Fortune Business Insights

India’s Sovereign Moment

Most nations that want to compete in advanced semiconductors face a painful reality: they are decades behind the established foundry ecosystems of Taiwan, South Korea, and the United States. Catching up on mature digital silicon nodes is an expensive, backward-looking game.

Despite that lag, we’re seeing early results nearing commercial scale from startups such as Mindgrove Technologies, another Speciale portfolio company, as India’s RISC-V ecosystem develops.

Neuromorphic computing, however, changes the equation. This is an emerging field where no country has yet established dominance—where India’s deep academic talent base and the forward-thinking capital commitment of the India Semiconductor Mission (Rs. 1,27,500 crore in expanded outlay) can position the country not as a follower, but as a founder of the next compute paradigm. The infrastructure for a genuine semiconductor renaissance is being assembled—and neuromorphic is where India can operate at the frontier, not the periphery.

The Academic Vanguard

Three of India’s premier research institutions are already producing world-class results—each on a distinct technical approach, together forming a credible foundation for an indigenous neuromorphic ecosystem.

BANGALORE · CENSE Indian Institute of Science Professor Sreetosh Goswami’s team engineered a molecular analog computing platform around a 60-nanometer ruthenium-based thin film capable of accessing 16,500 distinct conductance states—a breakthrough published in Nature. Delivering 4.1 TOPS/W of energy efficiency at a stable 14-bit analog resolution, the memristor eliminates the von Neumann bottleneck by computing directly inside memory. The team demonstrated its potency by recreating NASA’s Pillars of Creation image on a tabletop system—a task that would otherwise require a supercomputer.
16,500 conductance states	4.1 TOPS/W	14-bit resolution	60nm thin film

MUMBAI · MELODE LAB IIT Bombay Professors Udayan Ganguly and Maryam Shojaei Baghini’s lab has built the world’s first spiking neuro-synaptic core on mature 45nm Silicon-on-Insulator technology. By exploiting quantum tunnelling current near a transistor’s off-current level, their artificial neurons operate at current levels 10,000× smaller than conventional transistors—implementing a mixed-signal system-on-chip that executes real-time spoken word recognition modelled on the biological auditory cortex.
45nm SOI	10,000× lower current	Real-time speech recognition

NEW DELHI · CYRAN AI IIT Delhi Professor Manan Suri’s research group has validated 28nm FeFET-based in-memory computing systems and is translating that research into commercial products through Cyran AI Solutions—a startup developing hardware-software stacks that run spike-based models on resource-constrained drone and satellite payloads. This is edge AI operating precisely where neuromorphic advantages are most decisive: no connectivity, tight power budgets, real-time inference.
28nm FeFET	In-memory computing	Drone & satellite payloads

“Neuromorphic computing has had its fair share of unsolved challenges for over a decade. With this discovery, we have almost nailed the perfect system—a rare feat.” SREETOSH GOSWAMI, ASSISTANT PROFESSOR, CENSE · IISC BANGALORE

The Investment Thesis

These are not incremental improvements. They are proof that India’s domestic laboratories are generating commercially viable, defensible intellectual property at the very frontier of AI compute. The trajectory from academic breakthrough to commercial deployment is compressing—and the geopolitical tailwinds, a global race for sovereign AI infrastructure, compress it further.

We view neuromorphic computing not as a niche research area, but as one of the defining battlegrounds of the physical AI era. The countries and companies that establish leadership in brain-inspired compute today will set the terms for AI at the edge for the next two decades. Rather than running an expensive race to catch up on mature digital silicon, India has the rare chance to stake a founding position in the next generation of compute entirely.

India has the science, the policy intent, and an emerging commercial ecosystem. What comes next is capital and conviction—backing the ventures and researchers who will carry these breakthroughs from laboratory to global deployment. That is precisely the bet we are making.