Infrastructure / Edge AI hardware

Texas Instruments AM67A

Texas Instruments edge-AI vision SoC on a 16nm FinFET process, pairing a quad Arm Cortex-A53 at 1.4 GHz with dual C7x DSP cores and matrix multiply accelerators delivering 4 TOPS. It integrates a multi-camera image signal processor taking up to four CSI-2 inputs, PCIe Gen3 and Gigabit Ethernet with TSN, and supports up to 8GB of LPDDR4. It is the silicon behind the BeagleY-AI board.

Supported by TI's Processor SDK and Edge AI Studio. The openness axis weighs a public datasheet and released TIDL tooling against silicon and firmware supplied as binaries. Verified 2026-08-13 via the TI product page, the edgeai-tidl-tools repository and the BeagleY-AI board page.

Openness

3 high confidence

3.0

schematics: partial (reference designs (BeagleY-AI board files open))
toolchain: open (open SDK + TIDL tools + Edge AI Studio)
datasheets: public (Rev B)
blobs: required (proprietary silicon + TIDL firmware)
retail: distributor (active via TI and partners)

Public datasheet and open SDK/TIDL tools on GitHub, commercially active, but proprietary silicon and firmware keep it short of fully open hardware. TI publishes the "AM67x Processors datasheet (Rev. B)" openly and links Jacinto 7 LPDDR4 board design and layout guidelines rather than the design of any product board, and ships a FIRMWARE-BUILDER-AM67A alongside the Linux SDKs; edgeai-tidl-tools is a public repository.

https://www.ti.com/product/AM67A recorded 2026-08-13
status ACTIVE; "AM67x Processors datasheet (Rev. B)"; "Jacinto 7 LPDDR4 Board Design and Layout Guidelines (Rev. F)" and evaluation boards rather than product board files; Order now with evaluation boards "In stock"; FIRMWARE-BUILDER-AM67A and PROCESSOR-SDK Linux/Android SDKs
https://github.com/TexasInstruments/edgeai-tidl-tools recorded 2026-08-13
public "Edgeai TIDL Tools and Examples" repository, 736 commits, with a Supported Devices and Compatibility table and a LICENSE in-tree

Adoption

3 high confidence

3.0

ACTIVE production part available from TI and distributors, and the basis of the widely sold BeagleY-AI board, giving solid commercial and community reach.

https://www.beagleboard.org/boards/beagley-ai recorded 2026-08-13
"Texas Instruments AM67A Arm-based vision processor" as the BeagleY-AI processor, with the dual C7x DSP and MMA rated at 4 TOPs
https://www.ti.com/product/AM67A recorded 2026-08-13
status ACTIVE with evaluation boards "In stock" and an Order now path

Capability

4 high confidence

4.0

4 TOPS plus a strong vision/ISP pipeline and 8GB memory make it a capable dedicated edge-AI vision SoC. The part page reads "up to 4 TOPS total" from 2x C7x DSP with MMA, quad Cortex-A53 at up to 1.4GHz, max LPDDR4 size of 8GB and a 600MP/s ISP feeding four cameras. It sits a step above beagley-ai, the 4 TOPS board built on it, because the SoC supports 8GB where that board fixes 4GB.

https://www.ti.com/product/AM67A recorded 2026-08-13
"up to 4 TOPS total" from 2x C7x DSPs with MMA at up to 1.0GHz; quad-core Cortex-A53 at up to 1.4GHz; "Max LPDDR4 size of 8GB"; 600MP/s ISP

Verified 2026-08-13