Floadia announced that it had developed a unique flash memory that can store seven bits of data per cell (7bpc) for ten years at 150°C, the company said. That's much denser than today's leading flash memory for storage devices, like SSDs, that can only store four bits per cell (QLC).
Developing non-volatile memory for microcontrollers and other highly integrated devices has its perks. On the one hand, reliability and data retention time are an absolute must. On the other hand, storage capacity, density, performance and pricing are important too. Bringing all these features together is tricky, and at times companies have to invent rather interesting technologies.
Floadia's new tech is largely based on the company's Sonos flash memory that's designed for microcontrollers. In a bid to enhance memory and make it store 7bpc for ten years at high temperatures, Floadia had to redesign the structure of charge-trapping layers by introducing a proprietary Ono film and redesigning the cell structure and control network. The company said it could only retain 7 bits of data per cell for around 100 seconds without those enhancements.
Typical 3D QLC NAND flash memory stores 4-bits-per-cell and supports around 1,000 program/erase cycles at normal temperatures. Floadia's new memory can store more charges per cell and has a longer retention time than regular flash, meaning it can store data without power for longer periods of time.
But there's a catch. Floadia's Sonos is aimed at microcontrollers that don't need to store loads of data and that are made using thick (55nm ~ 350nm) and outdated process technologies. To that end, the storage density of the Sonos memory is considerably lower than modern 3D NAND memory. The company also hasn't shared performance expectations.
Floadia's new type of memory is designed for applications like computing in memory (CIM). In particular, Floadia intends to apply the new technology to an artificial intelligence (AI) accelerator that performs inference operations at 300 TOPS/W. That's significantly more efficient than competing solutions, like CPUs or GPUs. The AI chip stores neural network weights in non-volatile memory and executes many multiply-accumulate operations in parallel by passing a current through the memory array. The CIM AI accelerator can be used for edge applications.
Floadia yet has to reveal all peculiarities of its new flash memory and accelerators, but so far, it looks like the company has made a breakthrough in flash memory. Normally, Floadia licenses its flash technologies to interested parties. However, we don't know whether the technology can be applied to typical 3D NAND.