Audio
Wake-word detection and acoustic event classification for devices that need to listen continuously within a tight battery budget.
Computing
with sound.
Cortisonic builds machine-learning hardware from arrays of nanomechanical oscillators. The physics does the computation, so intelligence can run always-on at the edge for a fraction of the power of digital AI.
Up to 1000×
Lower power than conventional AI inference
nW–mW
Target inference regime for always-on operation
0
GPUs, DSPs or backpropagation required
Backed by
Supported by
Why now
Compute has a power problem.
From gigawatt data centres to smart watches that need charging every night, energy now limits what AI can do. At the edge, devices must listen, sense and respond continuously within tight budgets for battery, size and heat.
The cause is architectural. Conventional processors keep memory and computation apart, so data shuttles constantly between them, and every signal must be digitised before any computation can begin. Each of those steps costs energy.
Cortisonic takes a different path. Our chips compute with sound: physical neural networks built from nanomechanical oscillators on silicon. They draw on a decade of foundational physics research and are designed for established semiconductor and MEMS manufacturing.
Read about Cortisonic in Forbes Australia and Main Sequence’s investment announcement.
Technology
The physics does the computing.
A conventional AI pipeline digitises a signal, filters it, transforms it and runs it through a neural network, spending energy at every stage. Our chips let a network of coupled oscillators respond to the signal directly. Their physical response transforms the input into features that feed a lightweight readout or a digital neural network.
We call these chips Sonic Processing Units (SPUs).
Input
An audio, RF or sensor signal is coupled directly into the chip as mechanical motion.
Oscillator network
Coupled nanomechanical oscillators project the signal into a high-dimensional state with intrinsic nonlinearity and memory.
Readout
A lightweight, trainable output layer turns the lattice state into a classification or prediction.
High-dimensional projection
Thousands of mechanical modes expand the input space, the way hidden layers do in a deep network.
Built-in nonlinearity
Each oscillator follows a well-understood nonlinear equation of motion, so there are no activation functions to compute.
Memory from motion
The network keeps ringing after an input passes, so recent history shapes every output. This suits streaming and time-series data.
Engineered for scale
Precision engineering of coupling, damping and topology shape the network’s behaviour. Standard semiconductor and MEMS fabrication processes give reliable, low-cost manufacture at scale.
More hardware photos
Applications
Always-on intelligence where power is the constraint.
Our chips excel at time-varying signals, run at extremely low energy, and are naturally robust to electromagnetic interference. That makes them suited to devices that must listen, sense and decide without a data centre.
RF
Identify signal types in crowded radio environments, helping remote sensors filter traffic and respond when it matters.
Distributed sensing
Local inference in sensor networks and remote platforms, where size, thermal and power budgets rule out conventional processors.
Time-series prediction
Predict how a system will evolve from recent sensor data, supporting real-time monitoring and control in robots, drones and industrial equipment.
Research foundations
Built on a decade of chip-scale wave physics.
Cortisonic grew out of research showing that sound can be guided, switched and made to compute on a silicon chip, and that the nonlinear wave dynamics our networks rely on can be engineered and measured there. Those results, together with patented phononic circuit components, are the building blocks of our technology.
-
Preprint, 2026
arXiv:2603.22898
Solitary waves in a phononic integrated circuit
Acoustic dark solitons form in phononic waveguides from the balance of dispersion and mechanical Kerr nonlinearity, and we image them through hundreds of collisions. Our oscillator networks compute with the same nonlinearity.
-
Science, 2025
Science 390, 371 (2025)
Nonlinear wave dynamics on a chip
A chip-scale wave flume built from superfluid helium films and optomechanical readout, revealing wave steepening, shock fronts and solitary-wave fission. A different medium from our phononic chips, but the same nonlinear wave physics, measured at the nanoscale.
-
Preprint, 2025
arXiv:2509.11560
Nanomechanical error correction
Collective dynamics and redundancy give nanomechanical systems a form of intrinsic error correction, a key requirement for scalable physical computing.
-
Physical Review Applied, 2024
Phys. Rev. Applied 21, 054029
Acoustically driven single-frequency mechanical logic
Logic operations performed by nanomechanical elements driven at a single acoustic frequency, showing that vibrations can both carry and process information on a chip.
-
Applied Physics Letters, 2024
Appl. Phys. Lett. 124, 013504
Directional emission in an on-chip acoustic waveguide
Sound launched in a single chosen direction along an on-chip waveguide, a step towards integrated acoustic circuits.
-
Physical Review Applied, 2021
Phys. Rev. Applied 15, 054036
Tunneling of transverse acoustic waves on a silicon chip
Transverse acoustic waves tunnel across a barrier on a silicon chip, offering a new way to couple and route mechanical motion.
Team
The team behind Cortisonic.
Our team brings together expertise in nanomechanics, chip design, fabrication and software, with experience building and scaling technology.
Dr Glen Harris
Chief Executive Officer
Background in nanomechanics, quantum sensing and deep-tech R&D. Glen has spent a career translating complex physics into useful technology.
Dr Christopher Baker
Chief Scientific Officer
One of the original minds behind mechanical intelligence at Cortisonic. Chris leads the science that turns physics into practical, scalable hardware.
Dr Matt Reeves
Lead Scientist
Theoretical physicist specialising in nonlinear dynamics. Matt connects the physics of Cortisonic’s chips with machine learning, developing the models and software that guide their design.
Dr Tim Hirsch
Hardware Scientist
Experimental physicist with expertise in nanomechanics, MEMS and nanofabrication. Tim builds Cortisonic's hardware compute capabilities.
Dr Tina Jin
Hardware Specialist
Experimental physicist focused on nanomechanics, working on the fabrication and characterisation of Cortisonic's devices.
Dr Michael Harvey
Technology Advisor
Extensive experience building and scaling deep-tech startups across Australia, Silicon Valley and Europe.
Roadmap
A full-stack path from physics to product.
-
Emulator
for prototyping
-
SDK
for developers
-
Reference designs
for repeatable builds
-
MEMS fabrication
with foundry partners
-
Demonstrators
with industry
Get in touch
Work with us.
If you have a demanding edge workload, or want to invest in a new foundation for computing, we would like to hear from you.
Prefer email? info@cortisonic.com
Join us
We’re growing our team across physics, engineering and software. Get in touch about opportunities at Cortisonic.
Write to us about roles