• 沒有找到結果。

Chapter 2 Overview of Low Power Design and Leakage Control

2.4 System-Level Power Control

2.4 System-Level Power Control

The previous sections described many circuit-level threshold-voltage control techniques. However, with the progress of semiconductor technology and the trend of System-on-Chip (SoC), a system-level power and performance control methodology is more efficient. In this section, a power management unit (PMU) and its functionalities are introduced. Besides, the concept of Voltage Islands and a power control technique using status table is described.

2.4.1 Overview of PMU

Power management is a real-time technique to dynamically monitor and control power distribution and performance of a chip. Fig. 2.24 shows the block diagram of a possible PMU solution. It’s mainly consists of supply and body-bias voltage generators, device performance and thermal monitors, control logics, clock generators, and state machines. The voltage generators internally supply various voltages to serve as supply voltage or body bias; the monitors observe and detect the device performance and the temperature to keep the functionality and performance. Control logics are used to control the operation modes of functional blocks, and the power management state machine keeps tract of the mode transitions of functional blocks.

Fig. 2.24 A possible solution of PMU.

PMU monitors the power and performance conditions of all functional blocks, and executes operating mode transitions according to the activities. That is, if a functional block is in its high power and performance state but without any task to do, PMU sends control signals to change the power state. In addition, if the thermal detector observes that the temperature of a functional block goes too high, PMU enforces the IP to slow down the operating speed.

2.4.2 Concept of Voltage Islands

Voltage Islands are areas (logic and/or memory) on the same chip that are supplied by different voltage sources [2.20]. As discussed before, the various voltages may come from DC/DC converters. Voltage Islands restore the concept of individual voltage optimization of functional blocks to SoC design. Individual functional blocks of the SoC design can have different power characteristics from the rest of the design, and can be optimized accordingly. For example, the most performance-critical element of the design, such as a processor core, requires the highest voltage to maintain the required high performance. On the other hand, such as memory cells or control logics may not require this level of voltage. Therefore, significant power can be saved if they can run at lower voltages.

Fig. 2.25 Multiple on-chip Voltage Islands.

Fig. 2.25 shows the multiple on-chip Voltage Islands, which are operated under different supply voltages. In general, the circuits in the same Voltage Island have similar operation characteristics. As in Fig. 2.25, for instance, the Voltage Island 2 is a DSP (digital signal processing) processor and it can be fully shut down if we know that there are no DSP operations needed. An effective way is to add gating devices between the processor and the supply voltage or the ground.

Power domains are areas within an Island supplied by the same power supply but have distinct gating devices. According to the operating characteristics, part of an Island can be power gated but others are still power on.

2.4.3 Managing Threshold-Voltage Through a Status Table

A Vt management scheme using programmable status table is illustrated in Fig.

2.26 [2.21]. It consists of an instruction decoder, status monitor, programmable registers, and some logics. The decoder decodes an instructions and the requirement table identifies the function units that are required to execute this instruction. The status table stores the present power status of each function units. The logic observes whether the power requirement and power status are in agreement. If an instruction requires one function unit that is in lower power state, the execute logic sends control signals to change the power state to a higher level. A higher power state means both larger supply voltage and higher operating frequency. On the other hand, if another one function unit that is in higher power state is not required, the execute logic lowers

the supply voltage and frequency or even shuts down the function unit.

The existence of override register allows direct control of power and speed by application software. In other words, the override register can be programmed to directly control the power and speed of function units regardless of the internal managing scheme. For example, a reset signal from software can initialize the tables and put all the function units to the lowest power state.

The power latency table, as the broken-lined square in Fig. 2.26, is included but it does not appear in [2.21]. The power latency table contains the information of time periods required for power state transitions. The time latency of power state transitions must be taken into account since input data should be stalled until the required power state is ready. A false operation results from the execution under incorrect power state.

Fig. 2.26 A Vt management scheme using programmable status table.

2.4.4 Execution Circuit for Power and Performance Transition

Fig. 2.27 shows an execution circuit to adjust the voltage and frequency. Assume that each function unit has dedicated DC/DC converter and body bias generator, which can be independently controlled by PMU. The voltage generators can be shut down if the function unit is inactive. Moreover, the supply voltage and body bias can be dynamically adjusted if the DC/DC converter and body-bias generator are configurable.

Each function unit has a dedicated frequency divider to vary the operating frequency. According to the control signals from PMU, the frequency of function unit can be adjusted through a frequency divider. Undoubtedly, the clock signal fed into function unit can be gated for further power saving.

Fig. 2.27 Execution circuit to control power and frequency.