In Memory Digital Compute for AI Edge Inference.

Redefining Physical Intelligence.
The most efficient chip for edge devices, for applications constrained by power, latency, or data security & privacy.

Gen 1
Gen 1
Gen 1
Gen 1
Gen 1
Gen 1
Gen 1
Gen 1
Gen 1
Gen 1
Gen 1
Gen 1
Gen 1
Gen 1
Gen 1
Gen 1
Gen 1
Gen 1
Gen 1
Gen 1
Gen 1
Gen 1
Gen 1
Gen 1
Gen 1
Gen 1
Gen 1
Gen 1
Gen 1
Performance
0.2 TOPs · 4M parameters
Use Cases
Medical Devices · Wearables & Audio
Applications
Hearing aids · Smart earbuds · Health rings
Technology

Digital in-memory compute.

Bit-exact & deterministic

Every inference produces the same answer, every time. No analog drift, no spiking variance.

Memory and compute, fused

Weights live where the math happens. No external memory, no off-chip bandwidth bottleneck. Energy goes into computation.

Standard CMOS (7nm, 14nm, 22nm etc.)

Mature process. No exotic memristors, no custom fab. Predictable yield, predictable cost, predictable supply.

Modular by composition

One chiplet design. 1x for hearing aids. 4x for cobots. 8x for drones. Modular by composition.

Memory and compute, in the same place.

Conventional architectures spend most of their energy moving data between memory and compute. Digital in-memory compute eliminates this.

Traditional architecture
BIIO · Digital IMC
~0×
Less data-movement energy
vs. traditional von Neumann
~0×
More compute per watt
on edge-AI workloads
~0×
Faster on-chip access
vs. off-chip memory
Performance

Built and Spec'd for Edge.

Performance
0.2 TOPs
→ 60 TOPs
Scales linearly from a single tile to a 16-tile package. Same silicon, same toolchain.
On-die Parameters
4M
→ 1B
All weights stored on-die. Scalable as per needs.
Footprint
Wearable
→ Data Center
From milliwatt wearables to power-hungry infrastructure, one modular architecture fits every form factor.
Customization
Flexible
ML Stack
Deploy any model across all tile configs without retraining or recompilation. One software stack.
Team

Hardware isn't hard for this team.

Four core operators with deep silicon and physical-AI experience. Across prior tenures at AMD, ARM, Demant, and global research institutes, the team has shipped real silicon at scale.

5+
Chip designs shipped
30+
Hardware products in market
15+
Peer-reviewed publications
Demant
Demant
AMD
AMD
Intel
Intel
Microsoft
Microsoft
Amazon
Amazon
Tyndall
Tyndall
Danish Aerospace Company
Contact

Let's talk silicon.

For engineering partnerships, design-ins, or technical due diligence, get in touch directly.