TheLostSwede
News Editor
- Joined
- Nov 11, 2004
- Messages
- 16,056 (2.26/day)
- Location
- Sweden
System Name | Overlord Mk MLI |
---|---|
Processor | AMD Ryzen 7 7800X3D |
Motherboard | Gigabyte X670E Aorus Master |
Cooling | Noctua NH-D15 SE with offsets |
Memory | 32GB Team T-Create Expert DDR5 6000 MHz @ CL30-34-34-68 |
Video Card(s) | Gainward GeForce RTX 4080 Phantom GS |
Storage | 1TB Solidigm P44 Pro, 2 TB Corsair MP600 Pro, 2TB Kingston KC3000 |
Display(s) | Acer XV272K LVbmiipruzx 4K@160Hz |
Case | Fractal Design Torrent Compact |
Audio Device(s) | Corsair Virtuoso SE |
Power Supply | be quiet! Pure Power 12 M 850 W |
Mouse | Logitech G502 Lightspeed |
Keyboard | Corsair K70 Max |
Software | Windows 10 Pro |
Benchmark Scores | https://valid.x86.fr/5za05v |
Floadia Corporation, headquartered in Kodaira-shi, Tokyo, has developed a prototype 7-bit-per-cell flash memory chip that can retain analog data for 10 years at 150 degrees Celsius by devising a memory cell structure and control method. With the existing memory cell structure, the problem of characteristic change and variation due to charge leakage was significant, and the data retention was only about 100 seconds.
Floadia will apply the memory technology to a chip that realizes AI (artificial intelligence) inference operations with overwhelmingly low power consumption. This chip is based on an architecture called Computing in Memory (CiM), which stores neural network weights in non-volatile memory and executes a large number of multiply-accumulate calculations in parallel by passing current through the memory array. CiM is attracting worldwide attention as an AI accelerator for edge computing environments because it can read a large amount of data from memory and consumes much less power than conventional AI accelerators that perform multiply-accumulate calculations on CPUs and GPUs.
This memory technology is based on SONOS-type flash memory chips developed by Floadia for integration into microcontrollers and other devices. Floadia made numerous innovations such as optimizing the structure of charge-trapping layers, i.e. ONO film, to extend the data retention time when storing 7 bits of data. The combination of two cells can store up to 8 bits of neural network weights, and despite its small chip area, it can achieve a multiply-accumulate calculation performance of 300 TOPS/W, far exceeding that of existing AI accelerators.
View at TechPowerUp Main Site
Floadia will apply the memory technology to a chip that realizes AI (artificial intelligence) inference operations with overwhelmingly low power consumption. This chip is based on an architecture called Computing in Memory (CiM), which stores neural network weights in non-volatile memory and executes a large number of multiply-accumulate calculations in parallel by passing current through the memory array. CiM is attracting worldwide attention as an AI accelerator for edge computing environments because it can read a large amount of data from memory and consumes much less power than conventional AI accelerators that perform multiply-accumulate calculations on CPUs and GPUs.
This memory technology is based on SONOS-type flash memory chips developed by Floadia for integration into microcontrollers and other devices. Floadia made numerous innovations such as optimizing the structure of charge-trapping layers, i.e. ONO film, to extend the data retention time when storing 7 bits of data. The combination of two cells can store up to 8 bits of neural network weights, and despite its small chip area, it can achieve a multiply-accumulate calculation performance of 300 TOPS/W, far exceeding that of existing AI accelerators.
View at TechPowerUp Main Site