US 12,124,827 B1
Method and system to expand accessible on-chip memory (OCM) of an inference engine
Ulf Hanebutte, Gig Harbor, WA (US); Senad Durakovic, Palo Alto, CA (US); and Mohana Tandyala, Fremont, CA (US)
Assigned to Marvell Asia Pte Ltd, Singapore (SG)
Filed by Marvell Asia Pte Ltd, Singapore (SG)
Filed on Oct. 14, 2022, as Appl. No. 17/966,380.
Claims priority of provisional application 63/317,118, filed on Mar. 7, 2022.
Int. Cl. G06F 8/41 (2018.01); G06F 3/06 (2006.01); G06N 5/04 (2023.01)
CPC G06F 8/453 (2013.01) [G06F 3/061 (2013.01); G06F 3/0659 (2013.01); G06F 3/0673 (2013.01); G06N 5/04 (2013.01)] 20 Claims
OG exemplary drawing
 
1. A system comprising:
a first processing tile comprising at least a first processing element and a first on-chip memory (OCM), wherein the first OCM of the first processing tile is configured to receive external data for local storage and for local accessing and processing by at least the first processing element; and
a second processing tile comprising at least a second processing element and a second OCM,
wherein a portion of the external data is transmitted from the first processing tile to the second processing tile for temporary local storage in the second OCM of the second processing tile if an amount of the external data is greater than an amount of available space within the first OCM of the first processing tile, and
wherein the portion of the external data that is temporarily stored locally in the second OCM of the second processing tile is transmitted back to the first OCM of the first processing tile for local storage before the portion of the external data is needed by at least the first processing element.