Engineering note
FPGA, NPU or MPU? Partitioning an Edge AI Architecture
A practical guide to partitioning edge AI workloads across FPGA, NPU, MPU and MCU resources by latency, power and software needs.
Read the note →Modular compute
Embedded Compute Modules for Linux, RTOS and Edge AI
A scalable System-on-Module approach that lets programs select the right compute tier while reusing carrier design, software architecture and production infrastructure.
The best processor for one product may be excessive or insufficient for another. A modular compute family allows the same carrier philosophy, boot strategy and application interfaces to span deterministic control, industrial Linux and high-end heterogeneous workloads.
Alvexis develops compute modules and mission carriers together. The module concentrates processor, memory, storage and boot functions; the carrier implements power, physical interfaces, protection and program-specific connectors.
System architecture
1
Compute tier
MCU/NPU, industrial MPU or heterogeneous application processor.
2
Memory and boot
DDR, eMMC, QSPI and secure-boot architecture.
3
Module interfaces
High-speed expansion for Ethernet, CAN, USB, camera and GPIO.
4
Mission carrier
Power, protection, connectors, thermal path and production test.
Capabilities
Engineering scope
- Multiple performance and lifecycle tiers
- Embedded Linux, RTOS and mixed-criticality options
- Secure boot and signed-update roadmap
- GbE, CAN-FD, USB and camera interfaces
- Vision, industrial I/O and test carrier variants
- BSP, device tree, drivers and manufacturing support
Applications
Typical use cases
- OEM embedded products
- Camera and AI payloads
- Industrial gateways
- Rugged mission computers
- Data concentrators
- Long-lifecycle platform modernization
Specification discipline
What is defined during a pilot.
Values depend on the selected sensors, processor, interfaces, environment and acceptance method.
| Area | Definition |
|---|---|
| Tier selection | Based on workload, thermal budget, lifecycle and software ecosystem. |
| Module compatibility | Commonality is maximized without forcing incompatible high-speed constraints. |
| Security | Threat model and provisioning flow are defined per program. |
| Program status | Platform family under development with carrier variants planned around customer pilots. |
Related
Continue exploring.
Edge Platforms
Rugged Edge Computer
Rugged fanless edge computer with wide-input power, configurable I/O and embedded Linux or edge AI compute options for field deployment.
Explore system →
Edge Platforms
Verification & Data Logger
CAN, Ethernet and discrete-I/O verification logger for electronics bring-up, system testing, timestamped recording and fault analysis.
Explore system →Define the embedded compute module configuration.
Send the interfaces, environment, decision metric and pilot quantity. We will identify the architecture and validation plan.