Using its memory design expertise SureCore can help AI chip developers with complex power and performance challenges.
Commenting Paul Wells, CEO at sureCore, said, “AI can be a daunting new challenge for many chip developers. Because of the computational intensity AI chip designers need to pay special attention to power consumption. They can no longer make do with standard memory solutions and must innovate at the memory sub-system level to deliver lower and lower power targets.
“Our sureFIT design service is ideal for AI applications. We are excited that we can offer our many years of expertise designing for high performance, low power and low voltage memory to help AI customers.”
The sureFIT design service allows bespoke memory solutions to be created that are tailored to meet specific application needs.
Recently, the sureCore team has been working with a number of cutting-edge companies developing both Server-AI and Edge-AI System-on-Chips. In both cases the underlying memory architecture is increasingly being seen as a critical factor in delivering a competitive solution.
Raw performance or ultra-low voltage operation will help move the dial respectively in the Server or Edge spaces. However, to achieve the step change needed involves crafting a bespoke memory sub-system architecture including in-memory compute capabilities. This will reduce data traffic to and from the processor thereby increasing the efficiency of model execution.
Another area where sureCore’s engineering team can help is its deep understanding of the nature and importance of statistical analysis where large memory arrays are concerned. Many AI chips will integrate many millions of bit cells, and whether these are foundry designed or custom, it is essential that their behaviour at the operating extremes is well understood and characterized so as to guarantee at least six sigma performance. Only by doing so can competitive yields be achieved.
“AI chip developers demand ultra-low power memory IP, and the sureFIT design service can deliver on their complex design and PPA requirements,” said Wells.
