From Cloud to Sensor: Neuronova on Building Truly Efficient AI Sensors

As AI moves from the cloud to the edge, power is becoming the new bottleneck, replacing performance. Always-on sensing, privacy-first processing, and battery-limited devices are forcing us to rethink how and where intelligence should run. In the middle of all this, a new class of analog, neuromorphic architectures is quietly redefining what truly efficient AI looks like.

“Doing more with less” has been the motto of the Milan-based deeptech startup, Neuronova. Founded by a team of three PhD-level engineers, Neuronova’s neuromorphic chip processes AI workloads at less than 1µW using a fully analog pipeline from sensor to chip.

Neuronova is redefining computing by placing intelligence at the forefront and embedding it directly into the sensors. This shift turns intelligence into something distributed and energy-efficient, transforming every sensor into a tiny, aware computing unit. The result is a new computational layer for the physical world—one designed to power the trillion‑sensor future with meaningful awareness instead of just raw data.

We sat down with the Neuronova team to learn more about their neuromorphic chip, the company’s mission, and their collaboration with Cadence.

What Applications Is Neuronova Focusing On?

Our first focus is consumer audio and wearable electronics, devices that live close to the human body and sense the world continuously. Hearing aids, TWS earbuds, and smart glasses need to detect voice, ambient sound, and context while consuming only microwatts of power. 

In this framework, Neuronova’s analog neuromorphic processor enables on-device audio tasks such as keyword spotting, acoustic scene classification, voice activity and wake word detection  all running locally without cloud latency or privacy risks.

Beyond audio, the same platform extends naturally to health monitoring and motion intelligence, including anomaly detection in biosignals such as electrocardiograms and respiration.

In the medium term, our technology scales to industrial IoT, robotics, and automotive, where continuous perception and predictive analytics must operate without connectivity or power overhead.

What Are Some Challenges Neuronova Is Facing Today?

The issues and challenges we face today are closely linked to our customers' pain points. Device makers tell us that every new feature they add is constrained by a triangle of power, bill of materials, and complexity.

To bring serious neural networks to earbuds, smart glasses, or hearing aids, they must either keep the digital chain always awake or push processing to the cloud. Both options are expensive in battery, silicon area, radio usage, and engineering time. At the same time, they are under pressure to differentiate in very crowded markets without exploding costs or redesigning their whole architecture. Our challenge is to give them a new hardware option that fits into this reality, not as a research toy but as a production-grade platform that they can trust and scale.

We address this challenge through a very deliberate positioning. On the manufacturing side, we rely on a mature, cost-effective fabrication process and European silicon, which allows us to align with the strong price sensitivity of the sensor and component market. On the product side, our hardware–software co-design is focused on robustness and simplicity: both the hardware and the software are designed to integrate easily into existing architectures and standard workflows, behaving like a standard component while still enabling always-on intelligence below 1 µW.

How Were You Dealing with Them Before?

Before building this analog neuromorphic platform, the markets were dealing with these issues by squeezing incremental gains out of digital. They hand-tuned firmware, aggressively gated clocks, underclocked digital cores, and used simpler models than they actually wanted, just to stay within power and cost envelopes.

Some experimented with fixed-function analog front-ends or very narrow anomaly detectors, which helped in individual use cases but did not generalize and increased system complexity because each function needed its own chip or custom block. In short, the approach was a patchwork of optimization rather than a clean, scalable solution.

Why Choose to Improve Sensors?

The inspiration behind Neuronova is the belief that true intelligence should live where perception happens. Today’s computers think about data, but they don’t think with the sensor. Neuronova builds chips that work like the human brain, using continuous analog signals instead of digital bits, and networks of artificial neurons and synapses that process information through time rather than through precise arithmetic.

How Do You See Neuronova Scaling?

Neuronova scales along the same path as integration itself. The first phase, now underway, delivers discrete chips for wearables and hearables in the tens of millions of units.

The next step embeds our core into system-in-package (SiP) modules with microphones, IMUs, and biosensors, multiplying reachable volume to hundreds of millions of units through existing sensor vendors. The long-term evolution is full in-sensor integration, where our neural core becomes a standard feature of the sensor die, also licensed as IP and shipped in billions of devices every year.

As computing moves from the cloud to the edge, sensors will become the primary interface between the physical and digital worlds, and Neuronova will provide the intelligence layer, making analog AI as pervasive as MEMS technology is today.

What Has It Been Like to Partner with Cadence?

Neuronova’s collaboration with Cadence is built on Cadence’s commitment to supporting emerging companies by providing access to advanced design tools under favorable terms. These tools enable us to design and prototype silicon components and to model complex real-world behaviors that would be difficult to anticipate without detailed simulation.

Cadence has been a consistently reliable partner. The team in Milan provides steady support, clear guidance, and timely access to the information we need. Our primary contact is available whenever clarification is required.

Before selecting Cadence, we explored several options within the broader EDA ecosystem. Other platforms offered useful capabilities, but Cadence stood out for the maturity and cohesion of its environment, which matched the level of design work our projects demand.

How Are Cadence Tools Integrated in Neuronova’s Development Process?

Cadence’s software suite plays a central role in the development of Neuronova’s IP. It supports the design of our flagship H1 processor across the entire workflow, from the earliest cell-level concepts to the final signoff required for fabrication.

We rely on Virtuoso Studio for schematic creation, ultra-low-power circuit simulation, and the design of compact memory arrays. The Pegasus Verification System supports our DRC and LVS verification, helping streamline the signoff process. For the digital components that interface with our memory structures, we use Xcelium Logic Simulation for Verilog simulation, together with Genus Synthesis Solution and Innovus Implementation System for synthesis and place and route.

How Do You See Cadence and Neuronova’s Partnership Moving Forward?

What the future holds for Neuronova’s applications is a progressive shift from “smart devices” to truly perceptive environments. In the near term, our analog processor will become a standard building block in audio wearables, hearing aids, smart glasses, and health wearables, enabling continuous listening and monitoring at power levels compatible with tiny batteries and even energy harvesting.

Cadence and Neuronova together can accelerate and industrialize this shift. Cadence provides the tools, flows, and verification environments that turn neuromorphic ideas into manufacturable silicon and system designs, while Neuronova contributes a new class of analog neural architecture and use cases built around real sensors and real products.

Looking forward, this collaboration can define reference flows for analog AI inside sensors, standard-cell libraries, and IP blocks optimized for temporal neural processing, and complete reference designs for audio and health devices.

Cadence scales the impact of Neuronova’s technology from a single chip to an ecosystem, making analog intelligence a mainstream design option rather than a niche experiment and helping the entire industry move toward edge native, sensor-centric AI.