Bring Powerful AI to the Edge. Within the Power Budget.
Lower Power. Less Data Movement. Smaller Systems.
Edge AI operates under fundamentally different constraints. Battery-powered and power-constrained devices demand ultra-low energy consumption, minimal heat generation, compact form factors, and reduced dependence on costly external high-bandwidth memory.
The MLX200 and MLX300 series SoCs address these challenges at the architectural level. By storing neural-network weights in non-volatile memory and computing where the weights live, RRAM-based Analog In-Memory Computing dramatically reduces memory access and data movement — enabling powerful AI within the power, thermal, space, memory, and cost constraints of edge devices.
Built for energy-conscious innovation
Drastically lower power draw
Ideal for battery-powered and thermally constrained environments
Supports INT4 and INT8
Balances speed and precision to optimize performance per watt
Switchable modes
Toggle between ultra-low power and high-performance configurations
Breakthrough the Bottleneck
Traditional chips spend more time and energy moving data than processing it. MLX200 eliminates that inefficiency by combining compute and memory into one physical location. This architecture:
Remove the Memory Wall
Reduce latency
Slash energy consumption at every stage of processing
The result is a chip that does more with less—on the edge, where it matters most.
Built for edge AI seeking to move fast and stay cool
MLX200 delivers high-performance AI in power- and space-constrained environments, enabling real-time decisions without cloud latency or thermal issues.
AR/VR
Headsets
Delivers local AI processing for immersive, low-latency spatial interactions
Health
Monitoring
Continuously analyzes vitals on-device, extending battery life and preserving privacy.
Audio
Devices
Delivers accurate, always-on wake-word and sound detection directly on-device.
Process data right in the sensors
Reduces workload on the main processor, enabling slimmer designs, longer battery life, and all-day wearability without performance trade-offs.
AI for continuous vitals tracking
Analyzes biometric signals locally to extend battery life, enhance privacy, and remove reliance on phones or the cloud.
Low-power AI for voice recognition and sound processing
Reduces latency and energy use, enabling responsive, battery-efficient performance in earbuds, wearables, and smart speakers.
We are here to save the environment from AI
Meet the challenge of deploying an efficient edge AI solution by optimizing for power and latency on TetraMem’s world-class analog computing architecture
Average Energy Consuption (Lower is Better)
Why MLX200 Is Different
Analog in-memory compute
Matrix ops occur where data live, not across bus line
Massively parallel architecture
More compute throughput at lower energy cost
Event-triggered AI
Only wake full processing when needed
On-chip non-volatile memory
Store models, firmware, and logs internally
UNLOCKING
THE FUTURE
TetraMem enables edge devices to do more with less. By processing data directly in memory, we dramatically reduce power consumption and unlock real-time AI for products that couldn’t support it before.