Semiconductor strategies for low power consumption

Power consumption requirements for embedded systems, especially in automotive applications, are very stringent. Cars consume a lot of power while in full operation and also generate a lot of heat. The dissipation of that heat at high temperatures can become a serious issue, especially when a car is parked in the sun for an extended period. Furthermore, some control units must be capable of being operational even when the engine isn't running. These types of units drain the battery, so minimizing their power consumption is important.

There are multiple approaches a system designer can take to minimize vehicular power consumption. There are also things that semiconductor manufacturers can do to assist the designer.

Enhanced voltage regulators

While automotive systems usually work with voltages of 3.3 or 5 V (referred to as system voltage in this article), current embedded microcontrollers are designed to operate at 2.5 V or lower (referred to as core voltage). To be compatible with the system voltage without requiring two voltage supply lines on the board, state-of-the-art microcontrollers incorporate regulators that convert the system voltage to the core voltage.

Besides providing only a single power supply to the microcontroller, the on-chip voltage regulator also enables control based on the state of the microcontroller. The regulator is active during normal operation of the microcontroller, but it can be shut down automatically when the microcontroller is put into a low-power state, thereby reducing overall system power consumption and saving precious resources.

There are two main methodologies available for on-chip voltage regulators: linear and switching voltage. A linear voltage regulator works like an adjustable resistor in such a way that the output voltage is constant. A switching regulator periodically charges a capacitor through an inductor, using the rule that an inductor cannot change its current instantaneously.

While the advantage of a linear regulator lies in its simplicity and its quasi-static behavior if the output current is constant, the efficiency depends strongly on the required voltage drop and the output current. The higher either of these values, the higher the overall power wasted for unwanted heating purposes. If we assume a 3.3-V system voltage and a 2.5-V core voltage, 24 percent of the microcontroller's power consumption is used up in the voltage regulator. It gets worse if we assume a 3.3-V system voltage and a 1.5-V core voltage, where 54 percent of the power consumption is wasted. This is also a significant issue for chip and package designers when trying to dissipate that heat.

While the input and output current is the same for a linear voltage regulator, it is different for the switching type. Using the example above, if we have a 3.3-V system voltage and a 1.5-V core voltage, using a switching regulator reduces the loss in the regulator from 54 percent to around 10 percent. But this improvement comes with a price. The switch is usually operated at frequencies of around 1 MHz, which might require additional measures for EMI and does require the addition of an external inductor, increasing the overall costs of the system.

Shifting gears

The operating frequency of a device has a direct (most of the time linear) impact on the power consumption. The slower the clock, the lower the power consumption.

When selecting a microcontroller for a specific application, one criterion to consider is that it must be powerful enough to handle its tasks even under worst-case conditions. To assist the system designer in achieving low power consumption under normal operating conditions, microcontrollers allow the shifting of gears — in other words, very rapid clock speed changes. Furthermore, to make that change seamless, the CPU speeds usually can be selected independently from the clock for the peripherals, so that you do not need to adjust your baud rate generator.

Another important feature in power control is for the microcontroller to have various operating modes that enable you to either manually or automatically shut down modules not in use. Every module on a chip that gets a clock signal also takes its share in the power consumption, even if it is not used. Modern controllers have features that allow enabling or disabling of the clock supply for a module.

Multiple clock sources and power modes also enhance overall power control. Implementing power modes enables the software programmer to switch to low power consumption or standby power without taking care of each and every module individually, although there is certainly still the option to do so. The main clock unit of a controller usually includes a phase-locked loop and other components that use a remarkable share of the overall microcontroller power. Running on a subclock instead of the main clock makes it possible to shut down the complete block for main clock generation, which allows the CPU to continue executing its housekeeping functions.

To limit the leakage current in standby modes, multiple power domains on a single device are implemented. While this requires some enhanced skills for the semiconductor manufacturer, the results are worth the effort. For example, it is usually desirable for the system to have the I/O port powered all the time. However, the core components of the device, like the CPU core, flash memory, clock generators or general peripherals, are required only while the device is in its normal operating mode. Depending on the application, you might want to keep some RAM or specific timers such as a watch timer alive and active. In a device with multiple power domains, each of these areas would have its own power supply line. As result, the power supply can be cut for each domain not in use, reducing the standby current of that area to zero.

Process technology vs. power
Even these measures are not efficient enough for the challenges resulting from the introduction of smaller geometries in the semiconductor processes. Semiconductor companies are under constant pressure to fit an increasing number of transistors on a manufactured silicon wafer. Smaller technologies make that possible. Lower costs and higher operating frequencies also are important reasons behind the quest to conquer the deep-submicron world.

This is great news, but what does it mean for your battery while your car is parked at the airport?

On the most basic level, semiconductor devices operate by switching small transistors between active and inactive states to achieve a desired circuit operation. These small transistors have a specified current consumption in both the active and inactive states. The current consumption in the inactive state is typically called leakage current. For low-power modes that still provide power to modules, the lowest power consumption is realized when the largest number of transistors are inactive. The end goal of any low-power mode is to turn off as many transistor circuits as possible, so that only leakage current flows. Of course, the developer still must have enough functionality to return to normal operation or do other housekeeping functions.

As mentioned, the laws of physics give us the following challenge: As semiconductor technologies get smaller, the leakage current of a transistor becomes larger. Depending on the specific implementation, a semiconductor's lowest-power mode could consume more current than its larger technology predecessor. Although this leakage current is very small, the concern is valid as an increasing number of transistors are implemented in a device. This is one of the challenges semiconductor manufacturers and embedded application developers face when moving to smaller technologies.

The current consumption of a CMOS transistor circuit, regardless of the technology, is much larger in the active, switching state. Compared with the inactive state, this has a much greater impact in the overall current consumption of a silicon device. In contrast to the inactive state, the active state's current consumption generally decreases with technology advancements. As automotive and other embedded applications migrate to nanometer technology, designers can expect to benefit from lower, individual-transistor current consumption for normal operating modes of a device. If functionality is kept the same, your embedded devices will use less battery power and your automobile, less fuel.

Jens Eltze (cc.eltze@necelam.com) is staff technical application engineer and Ian Byers (cc.byers@necelam.com) is system design engineer at NEC Electronics America Inc. (Santa Clara, Calif.).