Microchip has introduced a single-core embedded microprocessor for edge machine learning tasks. The SAMA7G54 is an ultra-low-power CPU that runs up to 1 GHz with integrated imaging and audio subsystems, including a MIPI-CSI2 (D-PHY) camera interface making it an excellent fit for vision-based applications.
The SAMA7G54 is an Arm Cortex-A7 CPU that runs up to 1 GHz with a single core. Microchip says by using a single higher-performance core, this MPU consumes less power but can still handle image and audio processing tasks.
On-chip is 128 kilobytes of SRAM and 5 kilobytes of secure backup RAM. In addition, 80 kilobytes of maskable ROM is available for a secure boot loader. Support for external ROMs includes Quad-SPI, eMMC, and others.
For off-chip RAM, the SAMA7G54 supports up to 16 Gbit of (LP)DDR2 and (LP)DDR3 running as fast as 533 MHz. Other memory interfaces are available as well.
For vision, cameras work over a 12-bit parallel or MIPI-CSI2 interface. The SAMA7G54 can process still images up to 8 megapixels or 30 frames-per-second of 720p video. While for audio interfaces, there are (up to) four I2S interfaces, one SPDIF transmitter & receiver, and a 4-stereo channel sample rate converter.
Communication with the processor is possible over the dual Ethernet ports, three high-speed USB PHYs, six CAN-FD channels, and three SD card (or eMMC) interfaces.
All of this fits on a 14 by 14 millimeter TFBGA343 package!
Microchip also offers an evaluation kit that breaks out the various IO interfaces. Additionally, the board includes industry "standard" headers such as the Raspberry Pi's 40-pin GPIO header, a Pi CSI camera interface, an unpopulated pad for an ATWILC3000 WiFi/BT module, and two mkroBUS connectors. Finally, it provides a four gigabyte DDR3L chip, one gigabyte of Flash, and two EEPROMs.