We need a better way of verifying power of devices by accurately measuring usage models.


By Bhanu Kapoor
Today, in the era of IPs and SoCs, verification consumes up to 70% of the design effort. Most of the rest of the design effort is focused on iterations related to meeting the performance goals. We also hear that power has become the No. 1 design criteria for several categories of semiconductor devices, but we still lack well-defined, power-oriented methodologies and tools that ensure we meet power goals.

Power estimation is a tough problem. It’s tough because of difficulties in modeling power accurately. There are a few issues here, but the key one is the data dependency of power. Power consumption is a combination of leakage and dynamic power components in the CMOS process technology. Leakage power depends upon the device state and dynamic power depends upon transition among these states. As a result, both the components of power and the overall power are data-dependent.

A typical consumer semiconductor device operates in different modes. A cell phone is designed to be able to perform various functions such as voice calling, data downloads, Web browsing, e-mail, gaming, photography, music replay, and video recording. A typical application processor SoC for the cell phone is designed to optimize power taking all these modes into account while putting the device into various power states. The data characteristics for all of these functions are different and any reasonably accurate power verification approach should accurately simulate the device in these modes.

In functional verification, we develop testbenches and write testcases to verify the design for various usage scenarios exercising the corner cases in its applications. We need to put together a similar effort in verifying power by creating a power infrastructure that exercises the device by injecting data that reflect various usage modes of the device. Power verification is the goal here and functional verification could be a byproduct here. We can refer to this infrastructure as powerbenches. Applications drive various usage modes, which in turn determine the power states, and the data is driven to the device in these states. So the system knowledge is important here just as the system specification is important to the task of functional verification.

–Bhanu Kapoor is the president of Mimasic, a low-power consultancy.