Localized droop detection and response for heterogeneous SoCs.
Whether you serve the ADAS, PC, or networking market, chances are that your SoC is heterogeneous; containing general processors and application-specific accelerators. Your solution might have a systolic array for convolutions, a cluster of CPUs for application code, or a look-aside crypto engine for packet security. While application-specific accelerators significantly improve performance and power efficiency, they can face localized droops from dynamic workloads or induce droops under heavy load. A localized voltage droop occurs when there is a sudden rush of switching activity driven by fluctuating workloads, which creates a drop in the supply VDD and might raise the ground voltage level (also known as a ground bounce).
A localized voltage droop might result in setup and hold time violations causing transient glitches and potentially catastrophic missionmode failures.
System architects can now mitigate droops with integrated droop response systems, which adapt dynamically to voltage droops (IR drops). The technique requires a programmable clock and a tightly coupled droop detector to create a fully self-contained droop response system.
Read more here.
 |
 |
 |
 |
 |
 |
 |
 |
 |  |
Leave a Reply