Google Coral Expansion Card
The Google Coral Expansion Card is a community-designed expansion card that integrates a Google Coral Edge TPU accelerator module into the Framework Laptop's expansion card form factor1. Created by Evgeni Genchev in May 2022, the project enables on-device machine learning inference (TensorFlow Lite) through a dedicated TPU chip housed in a standard expansion card slot12.
Background
The Google Coral Edge TPU is a small ASIC designed to accelerate TensorFlow Lite machine learning inference at the edge. It is used for computer vision, audio recognition, and other ML workloads, offering significantly better performance-per-watt than running inference on a general-purpose CPU13. The Coral ecosystem includes USB accelerators, M.2 modules, and bare accelerator modules (the 120-pin LGA chip used in this project)3.
Framework VP nrp (Kieran Levin) commented that a TPU expansion card had been on Framework's "wishlist of interesting projects"1.
Design
Accelerator Module
The Google Coral Accelerator Module is a 120-pin LGA chip measuring approximately 15 mm × 10 mm12. Despite the pin count, the design is simpler than it appears:
- 104 of the 120 pins are either 3.3V or GND
- Only 7 of the remaining 16 pins are required for basic operation via USB 2.0
- The module also supports PCIe Gen 2 x1 and USB 3.0 interfaces (the latter requiring special design considerations and Coral sales contact)13
PCB
Genchev designed the PCB in EasyEDA using the publicly available schematics from both Google Coral and Framework's expansion card reference designs12. The initial version uses USB 2.0, described by Genchev as "the easiest to make into a functioning circuit"1.
The design was completed before Genchev owned a Framework laptop — he developed it as a student unable to afford the hardware1.
Technical Specifications
| Specification | Details |
|---|---|
| TPU | Google Coral Accelerator Module (Edge TPU)3 |
| Interface | USB 2.0 (initial design)1 |
| Alternative Interfaces | PCIe Gen 2 x1, USB 3.0 (requires contact with Coral sales)13 |
| Power Consumption | Up to 2 W4 |
| Pin Count | 120-pin LGA (7 active signal pins for USB 2.0)1 |
| Module Size | ~15 mm × 10 mm12 |
| Source | EvgeniGenchev/Google-Coral-Expansion-Card (GitHub)2 |
| Design Tool | EasyEDA2 |
Development Timeline
| Date | Event |
|---|---|
| May 2022 | Genchev announces the design and publishes PCB files on GitHub1 |
| 2022–2023 | Global chip shortage prevents sourcing Coral modules; Mouser backorders estimated 6+ months out1 |
| Feb 2023 | Mouser lists 4,075 modules on order for July 2023 arrival1 |
| May 2023 | Community member chromakode successfully uses a Coral M.2 module in the Framework mainboard's WiFi slot as an alternative1 |
| Feb 2024 | Josh Cook reports successfully testing Genchev's design with a sourced module, considers build-to-order sales at $70–90 USD1 |
Alternative: Coral M.2 in WiFi Slot
A simpler alternative to the expansion card is installing a Coral M.2 Accelerator directly into the Framework mainboard's M.2 E-key WiFi slot14. Community member chromakode demonstrated this approach in May 2023, reporting that the single TPU M.2 module works in the WiFi slot1.
- The Framework 13 WiFi slot provides only a single PCIe x1 lane, not the dual lanes required by the dual-TPU M.2 variant
- Using the Coral M.2 displaces the WiFi module, requiring a USB WiFi adapter for wireless connectivity
- This is a trade-off between ML acceleration and internal WiFi
Use Cases
- Frigate NVR — Community member Blake Blackshear (creator of Frigate, an open-source NVR) expressed interest in using a Coral card for real-time object detection on Framework hardware1
- On-device ML inference — TensorFlow Lite model execution for computer vision, audio classification, and keyword spotting without cloud dependency
- ML development — Local testing and iteration of edge ML models
Related Projects
- TPU Expansion Card — Earlier community discussion (November 2021) proposing a Coral TPU expansion card, which led to Genchev's design4
- Google Coral USB Accelerator — The commercial USB dongle version that inspired this project3
- Coral M.2 Accelerator — The M.2 module that can be used directly in the Framework WiFi slot5