Low complexity, real-time adjusted power management policy using Golden Section Search

Golden Section Search (GSS) is applied to optimize the time threshold in a time-out power management policy. To reduce power consumption, modern system on chips selectively turns off some of its idling parts. If an idling block is kept awake for too long, it is wasting power; if it fails to predict a request, it will reduce the system performance due to wake-up time. The request of each part depends on external random events. This problem is usually addressed with a time-out policy, where the time-out value must be optimized in real time to cope with the system usage characteristics. Without the aid of prior knowledge, the GSS driven power policy consumed 15% more power than the absolute lower bound and 7% more power than the best fixed time threshold with a priori knowledge. The proposed method was evaluated with interrupt data of a 2G modem, captured from a real device in use.