Industry
High Performance Computing
Source: Techreport.com
In high-performance computing applications, computing power (and chip cost) is closely related to the amount of on-chip memory. High-end, server-class processors use increasingly large on-chip SRAM to reduce cache misses and avoid off-chip data accesses. SRAM is expensive, requires six transistors per cell, and often takes up most of the chip area. Chip manufacturers have taken extreme measures (such as using embedded DRAM) to reduce area, however, this often increases system complexity and power consumption.
Case Study: VC-MRAM vs SRAM
Technical Path |
area* |
Energy consumption |
Operating frequency |
Durability |
Zero standby energy consumption |
SRAM |
0.13um^2 |
<10 fJ |
1ns |
10^15 |
no |
VCMRAM |
0.034um^2 |
<10 fJ |
2ns |
10^14 |
yes |
*28 nm
Compared with SRAM cache, VC-MRAM storage density is increased by 3~4 times at the same technology node. It is worth noting that VC-MRAM is compatible with semiconductor processes, so many products that have begun to develop it as a replacement for SRAM are underway. In addition, as a non-volatile memory, it ensures the security of critical data when power is off.