Startup Putting the Artificial Intelligence Core in Solid-State Drives

Startup InnoGrit has debuted a set of 3 controllers for the solid-state drives (SSDs), which also includes one for the data centers embedding a neural network accelerator. They have entered a crowded market having claims of the advantages of power and performance over other rivals.

The chips have emerged as the NAND flash prices are indicating signs of the bottoming out. OEMs as well as the data centers have been expected to make use of the lower prices for continuing the shift from the hard-disk storages in notebooks as well as servers which require advantages of power, performance, or size.

In centers of data, the SSDs’ adoption is happening speedily, particularly with the price drops of last year.

At the high end, InnoGrit’s Tacoma makes use of four PCI Gen4 interfaces for supporting 16 NAND channels, and delivering about 1.5 million I/O operations/second (IOPS) at less than a 5W peak power.

Wu’s chief Executive said that having this combination, they could do intelligent processing like data labeling, utilizing the tool chain which was offered by Nvidia, or also some of the savvy customers would put in their own firmware.

In autumn last year, Samsung announced the launch of an SSD which embeds a Xilinx Zynq FPGA for handling the processing for a variety of Artificial Intelligence, database, as well as video apps.

The Tacoma makes use of a 64+8-bit data bus, and also a DDR3/4 as well as LPDDR3/4 DRAMs, and further provides AES-256, SHA3, and ECC security. The mid-tier controller or Rainier is a smaller version of their design, which supports an eight NAND channels as well as 32- and 16-bit data buses for the low-end servers or the high-end notebooks.

Rainier is delivering almost one million IOPS at less than a 3W peak power. Both the Rainier and Tacoma are supporting sequential reads at about 7 GBs per second and also sequential writes at 6.1 GBs per second.