The company claims to have a number of foundries capable of manufacturing its technology at various nodes as well as licensees developing products for those processes.
Now the company is demonstrating engineered memory arrays on test chips. Although the ReRAM is used to record 1s and 0s as any other non-volatile memory the arrays have been specifically designed to support classification applications such as facial recognition and number plate recognition.
Sylvain Dubois, vice president of business development at Crossbar, told eeNews Europe that the arrays have been made extremely wide to allow data from thousands of images to be brought in at the same time. "It is not only a memory array. It is array architecture with some logic integrated. There are algorithms embedded in the array. It works with specialist logic and beats any CPU/GPU architecture."
The ReRAM test chip demonstrations for AI-at-the-Edge showing simultaneous object classification in one iteration for facial recognition and car license plate recognition. The technology will be featured at the Embedded Vision Summit coming up in Santa Clara, Calif., May 21 to 24.
Dubois added that Crossbar is also working on using the analog properties of ReRAM memory element to create neuromorphic computing that is a closer model of brain activity.
Dubois did not reveal the process node that the test chips have been worked in, saying it was not relevant. "We want to use the test chips to persuade customers to go to 1X-nm. Dubois said much of Crossbar's development work was now targeting sub-20nm manufacturing.
Next: In pre-production