Architecture · A-Cube
Not an AI chip with a legacy brain, a new architecture built around what AI actually does. Energy-aware at every layer, scaling from coin-cell AI to Edge AI boxes
The PROBLEM
The multiply was never the expensive part. Moving the data was.
A-Cube
We put the compute inside the memory. The commute disappears. Compute where the data lives.
95%
of a neural net is matrix math. ~75% of the effort is moving it around.

Where compute happens
Inside Memory
Data Movement
Minimal
Efforts spent on moving data
Low
Efficiency
High
Legacy
Compute and memory sit apart. The chip spends its life shuttling numbers, not crunching them. ~75% of operations are just memory traffic.
Compute
Memory


Where compute happens
Outside Memory
Data Movement
High
Efforts spent on moving data
High
Efficiency
Low



Three breakthroughs as one: a physics-based brain, a responsive nervous system, and familiar language. Server-class AI with low power.
The Brain That think
CubicCore™
Server-class math. Microwatt power.
It runs the math on physics itself. Ohm's law multiplies, Kirchhoff's law sums - right inside the memory, almost for free. Digital keeps every result exact.
+
In-memory analog compute - no data commute
+
~1/100th the energy per operation vs. digital
+
Scales by replication - tile in more cores, rom a smart ring to autonomous robots
+
Fully programmable · 4–32-bit precision, tuned at runtime
+
Built in standard CMOS - no exotic process
Inside Sensemesh
Always-on inference at microwatts. Full power the moment it’s needed. Then straight back to idle, no reset, no wake-up.




ReflexSurge Mode
on-demand. high performance
SUBCONSCIOUS MODE
Always on. Ultra low power
Subconscious AI
+
Legacy chips switch off. A-Cube sleeps like you: the brain stays aware. A-Cube runs AI at microwatts while the host controller is powered down. Always processing, never draining.
Always processing, never draining.
ReflexSurge mode
+
When something matters, the brain wakes instantly. The moment SenseMesh flags a real event, runtime DVFS ramps the chip from subconscious idle to full performance
live, no reset — settles back down.
+
Once the task is completed, the chip settles back into subconscious mode.
A unified architecture for seamless adoption and scalability.
One architecture, one software stack, from a wrist to an autonomous drone.
GPX10 PRO
EDGE SENSOR
GPX40
CONNECTED VISION
GPX80
PERCEPTION & CONTROL
GPX Ultra
AUTONOMOUS AI
The efficiency gap isn’t a few percent. It’s a different category.
Every other approach trades something away. A-Cube doesn’t.
Rank
The Trade - off
1

A cube
Analog digital in memory
None
Power efficiency
~30 TOPS/W
Compute density
~5 TOPS/mm2
2

Flash
Compute in memory
Model Lock In
Power efficiency
~5 TOPS/W
Compute density
~1 TOPS/mm2
3

GPU
Paralle SMO
Too hot & power hungry
Power efficiency
1 TOPS/W
Compute density
~0.5 TOPS/mm2
4

MCU
Von-Neumann
Not built for AI
Power efficiency
0.1 TOPS/W
Compute density
~0.1 TOPS/mm2
A-Cube is live
Proven in silicon, shipping today.
A-Cube powers the GPX10 Pro's full neural networks on-device, at microwatt power.
512 GOPS
PEAK PERFORMANCE
High AI throughput enough to run full neural networks on-device, not rule-based shortcuts.

~80 uW
ALWAYS-ON POWER
Always-on AI inside a coin-cell power budget for months of battery, not hours.
A new architecture for AI, energy-aware at every layer, scaling from coin cell to cloud.
Whether you’re scoping a product or pressure-testing the architecture, there’s a direct path forward.
Explore how Ambient AI can unlock new capabilities in your wearable product. Strategic planning session with our applications team.
Buy an Evaluation kit.Be the first to access our upcoming Vision, Sound, and Industrial modules.
Talk to SalesPut A-Cube to Work
Whether you’re scoping a product or pressure-testing the architecture, there’s a direct path forward.
Get anEvaluation Kit.
Explore how Ambient AI can unlock new capabilities in your wearable product. Strategic planning session with our applications team.
Buy an Evaluation kit.Scale to increasethe volume.
Be the first to access our upcoming Vision, Sound, and Industrial modules.
Talk to Sales























