In the video above, Microchip’s Brad Poole, Edge AI Business Development Manager, and Bill Li, Applications Manager for Microprocessors, highlight the SAM9x75 and SAMA7D65 application processors in a simulated smart-home environment. In this case, the SAM9x75 handles the image capture and optimization with its image sensor controller (ISC). The image is then passed via Ethernet to the SAMA7D65 that runs the facial recognition stack (Fig. 1).

The SAM9x75 is based on an Arm ARM926 core. It incorporates 64K SRAM and a DDR2/3L interface. However, for this application, it’s the MIPI 4-lane CSI-2 camera interface and the built-in image sensor controller (ISC) that are important. The ISC supports ITU-R BT 601/656/1120 video interface up to 5 Mpixels as well as raw Bayer 12, YCbCr, monochrome and JPEG compressed sensors up to 12 bits. This provides a smart camera via Ethernet that improves the image quality before passing it on.

The SAMA7D65 is an Arm Cortex-A7 application processor running Linux. It gets the video stream via its 100/1000 Ethernet MAC with time-sensitive networking (TSN) support. The chip is supported by the Microchip Graphics Suite and MPLAB Machine Learning Development Suite. The chip is where the facial recognition software runs.

The application was running on the SAMA7D65 Curiosity Kit (Fig. 2). The board has a microSD slot plus 8 Gb of DDR3L, 64 Mb of QSPI NOR flash, 4 Gb of SLC NAND flash and a 2-kb serial EEPROM.

The SAMA7D65 Curiosity Kit provides two Ethernet ports (Fig. 3). One via a conventional RJ-45 jack and the other via the SODIMM socket. There are three on-board CAN-FD transceivers, too.

Also in the mix are a 40-pin Raspberry Pi expansion header, a pair of mikroBUS connectors, and a pair of PIOBU/system headers. The security support starts with a physically unclonable function (PUF) and includes secure boot, secure key storage, and high-performance crypto accelerators for AES, SHA, RSA, and ECC.