NPKyber
— From Sensing to Hardware Artificial Intelligence, an idea crystallising
NPKyber builds a modular path from signals to hybrid analog-digital AI.
Pre-launch, fast prototypes, honest about unknowns.
Project KyberGenesis:
-
KyberAcquire Kit:
Starting with plug-in sensor/interface modules and robotics kit
-
KyberGeneNet Studio:
Our tool for learning neural networks directly in circuitry
-
KyberFoundry:
Longer term, Kyber Foundry for a data-to-chip pipeline in Australia.
If you’re a maker, researcher, or partner, join the journey: KyberPartner
Why we exist
​
Modern AI is powerful but it wasn’t built for the physical world’s constraints. We see four gaps that block the next wave of artificial intelligence.
​
-
No built-in dynamics or memory:
Most deployed models treat time as an afterthought. Neurons and calssical ANNs don’t carry internal state.
​​
-
Inefficient density:
Throwing more parameters and FLOPs at every task wastes power, adds latency, and inflates BOMs. Especially when the job is simple (detect, threshold, stabilize, control).
-
Underused hardware primitives:
Today’s stacks ignore what hybrid analog-digital electronics already do well: sense, filter, compress, and respond with micro-watts and microseconds.
​
-
Sovereign capability matters:
Australia needs hands-on pathways, from education to fabrication, to design, build, and deploy intelligent hardware locally.
​
​
​Our approach
​
-
Neural networks in circuitry:
Learn compact analog-digital networks with intrinsic state and dynamics at the cell and regional level.
-
Right-sized intelligence:
Replace bulky MCU/CPU loops where possible with task-specific hardware behaviors that cut power and latency while improving robustness.
-
Leverage what exists:
Stand on proven semiconductor and computing tech—sensors, hybrid analog-digital interface, FPGAs/ASIC flows—guided by non-differentiable optimization.
-
Open where it helps, focused where it counts:
Open interfaces, formats, and learning content to grow the ecosystem; protect the synthesis engine and advanced IP to sustain the mission.
-
Maker-to-silicon pathway:
Start with modular kits and education, scale to a GPU-accelerated synthesis tool, and ultimately enable a data→design→hardware pipeline.
-
Sovereign capabilities:
"Lets bring em home boys and girls!​"
What this unlocks
​
Our belief is that networks with neuron dynamics and internal memory can do more with less than dense, stateless CNN/Transformer stacks.
Adding real dynamics and memory to neural architectures unlocks smaller, more efficient edge intelligence.
Emerging research in state-space, liquid, and continuous-time models and in function-on-edge networks shows we can match or exceed performance with fewer parameters and lower power, especially on temporal, sensor-driven tasks.
​
-
Lower power, lower latency, higher reliability at the edge
-
Simpler, cheaper devices for real-world tasks
-
Doing more with less by using existing analog/DSP blocks and commodity parts while uniting AI with electronics and biology-inspired dynamics.
-
A local skills and supply chain that can build and ship intelligent hardware from Australia
​
​
​
This is the problem we’re here to solve and the roadmap we’re executing to solve it!
Our Innovations
Leading the Charge
Our work encompasses a vast array of research and development, focusing on semiconductor manufacturing and advanced AI algorithms. We strive to stay at the forefront of technology, ensuring meaningful contributions to the tech community.
AI
Advanced Solutions
We’re developing an AI synthesis engine that turns real-world data into compact mixed-signal neural networks you can simulate or export for hardware implementation. Designed for low power, low latency, and robust control at the edge, with developer-friendly interfaces and workflows (licensing details to be announced).
Sensors
Precision Engineering
We build bio-inspired, mixed-signal sensor front-ends that fuse super-sampling, sensor fusion, and analog dynamics for cleaner, lower-power edge signals. An open sensor interface plus neuromorphic pre-processing outputs compact, training-ready features.
Data
Infrastructure Development
We build data pipelines for the edge—from RC capture and control logs to clean schemas, versioned datasets, and reproducible training/export workflows. Open formats, offline-first tools, and opt-in telemetry make integration simple while keeping your data portable and under your control.
Electronics
Creative Solutions
We design mixed-signal building blocks—modular analog/digital circuits and learnable “neural” cells—built for clarity, robustness, and low power with clean headers and PCB/VHDL exports.
Aspirational roadmap: a data→chip path (MPW shuttles and on-shore fabrication in Australia) to take proven modules into silicon.