Samsung Post Linux Patches For Trinity NPU Accelerator Driver

Samsung has posted a set of Linux kernel patches for enabling their Trinity neural processing unit (NPU) hardware as accelerators found within some of their embedded systems. Samsung Trinity NPUs have been in use since 2018 and their newer Trinity "TRIV2" accelerator is found within the latest Samsung TVs for AI purposes. Samsung hopes to upstream this new Linux driver simply named "trinity" into the mainline kernel.

The Samsung "trinity" driver in its current form is targeting the Trinity Vision 2 "TRIV2" hardware. Trinity Vision 2 supports accelerating image inference for CNNs using a deep learning accelerator while the general neural network layers are executed by a DSP. Trinity is made up of the deep learning accelerator, DSP, and a control processor. From the Linux kernel perspective, it's mainly about managing the control processor (CP) to then deal with the other hardware blocks. Trinity makes use of Arm's Advanced Micro-controller Bus Architecture (AMBA) standard.

The Linux kernel could soon support the neural network accelerator hardware found within Samsung's newest TVs for on-device AI.

With 2022 Samsung TVs, the Trinity TRIV2 can be used by AI-based applications for image recognition, picture quality improvements, and similar tasks. Linux user-space can access the TRIV2 with the kernel driver via GStreamer and its NNStreamer neural network plug-ins.

Those interested in Samsung's work on this Trinity NPU Linux kernel driver can check out the kernel mailing list for the initial details on this proposed new ~8.6k line driver.