This page is under construction.
DX-BItcELL technology
Will revolutionize AI systems with highly efficient
|
Rangduru's DX-bitcell is a memory bitcell architecture. It is a platform solution
for making denser emerging NVMs (non-volatile memory) that will displace DRAM and serve as the best working memory for AI. DX-bitcell consists of two elements: (1) a novel DX-transistor as a selector and (2) a storage element of whatever sort already produced by fab foundries for emerging NVMs. The DX-transistor is based on a half of a MOS transistor (a 0.5 transistor). Despite this smaller transistor, DX-transistor can drive a higher current for minimizing the area cost of a bitcell and enhancing the speed of bitcell operation. The DX-transistor can be formed on a standard logic process with the sole difference that an additional mask is required in the FEOL (front end of line) processing. DX-bitcell memories, such as DX-MRAM, can therefore be contrasted with DRAM, which requires use of a non-standard process that is costly and limits scalability. Whereas the storage element and logic gates in DX-bitcell memory can be scaled down to 3nm, DRAM faces scaling limitations somewhere around 10nm. Moreover, DX-bitcell memory can easily integrate with CPUs/GPUs/ALUs/glue-logic unlike embedded DRAM. DX-bitcell technology is applicable for multiple kinds of emerging NVMs such as MRAM (magnetoresistive random-access memory), PCRAM (phase-change random access memory) and RRAM (resistive random-access memory). DX-bitcell memory such as DX-MRAM, DX-PCRAM and DX-RRAM can serve as discrete or embedded products, enabling highly efficient in/near memory computing in AI systems. |
Bitcell Architecture comparison :
|
|
|
|
