What matters most when choosing an edge AI module?

Start with the workload, not the chip. Define your model, input resolution, and required frame rate first — that sets your TOPS band. Then constrain by power and thermal budget, pick an OS model, and finally weigh long-term supply and vendor lock-in. Raw TOPS is the last thing to optimise, not the first.

Should I pick a module by TOPS?

No. TOPS is a peak number that ignores utilisation, memory bandwidth, and model fit. Two modules with the same TOPS can differ several-fold on real frame rate. Size from your actual model and frame-rate target, then confirm with a benchmark on the hardware.

How important is long-term availability?

For any product shipping for years it is decisive. A module that disappears or a vendor you can’t second-source forces a costly redesign. Favour a written supply commitment and, ideally, an open standard you can re-source across vendors.

§ Guide

How to choose an edge AI module.

Picking edge AI hardware is mostly about getting the order of decisions right. Work the workload first and the chip last, and most of the choice makes itself.

The mistake most teams make is starting with a TOPS number or a familiar brand. Work these seven steps in order instead.

1. Define the model. Which network, at what input resolution, quantised how (INT8, FP16)? The model sets everything downstream.
2. Set the frame rate. Real-time control needs 30+ FPS; inspection might need 5. Multiply streams × resolution × FPS — that is your real compute target. Seehow many TOPS you need.
3. Fix the power budget. Battery, solar, or fanless designs live or die onTOPS per watt, not peak TOPS.
4. Check thermals. Can the enclosure shed the heat passively, or do you need active cooling? This often caps usable performance below the datasheet.
5. Choose the OS model. Linux, an RTOS, bare-metal, or a mix — seeZephyr vs Yocto vs bare-metal.
6. Confirm the lifecycle. How long is the part guaranteed available? Get it in writing — seeour 10-year commitment.
7. Avoid lock-in. Can you second-source the module and keep your firmware? Anopen standardlets you re-source across vendors without a respin.

How E1M maps to these steps

TheE1M lineup spans 0.25–33 TOPS across two footprints, so you can match step 2 without changing platform. OneSDK covers step 5 across every vendor, and theopen standard plus a written supply commitment cover steps 6 and 7.

Frequently asked questions

What matters most when choosing an edge AI module?: Start with the workload, not the chip. Define your model, input resolution, and required frame rate first — that sets your TOPS band. Then constrain by power and thermal budget, pick an OS model, and finally weigh long-term supply and vendor lock-in. Raw TOPS is the last thing to optimise, not the first.
Should I pick a module by TOPS?: No. TOPS is a peak number that ignores utilisation, memory bandwidth, and model fit. Two modules with the same TOPS can differ several-fold on real frame rate. Size from your actual model and frame-rate target, then confirm with a benchmark on the hardware.
How important is long-term availability?: For any product shipping for years it is decisive. A module that disappears or a vendor you can’t second-source forces a costly redesign. Favour a written supply commitment and, ideally, an open standard you can re-source across vendors.

Ready to build on an open edge AI standard?

Request a Dev Kit Browse E1M modules