Extending battery life with dual processors

There is no doubt that there is a trend towards lower power consumption and higher performance battery-powered systems. For battery power, consumers want portable electronic products to be able to "do more with less power", and many industrial products are also beginning to turn to battery power.

Digital signal processors (DSPs) are often used in applications that require high performance, resulting in ever-increasing clock speeds. Any processor that needs to run at high speeds and integrate millions of transistors needs to consume as little power as possible.

Over the past decade, architectural innovations and low-power strategies have driven DSP processing power to continue to increase, MIPS performance has continued to increase, and other performance parameters have also been rapidly improving. In this context, the battery life of DSP-based systems such as mobile phones and ultra-portable audio players has generally increased steadily and significantly due to these higher-performance DSPs.

Headphones and other portable consumer audio devices are important examples of devices that require optimal power efficiency. In addition, industrial and medical products are also moving towards highly integrated portable devices that support advanced functions. The power requirements of these devices are largely similar to those of ultra-portable consumer electronics.

As DSP technology has made rapid progress, microcontroller (MCU) systems are under pressure to reduce power consumption. Compared with DSP, MCU has its own unique advantages in reducing power consumption, such as fewer transistors, lower clock speed, and usually lower operating voltage.

Unlike DSPs that are calculated in microamperes, the standby current of power-optimized modern MCUs can be measured in nanoamperes. Although DSPs have greatly exceeded the performance of MCUs and have been significantly optimized in terms of power, there is a limit to what chip designers can do to save power.

Conventional knowledge

When it comes to extending battery life, common sense among system design engineers is that “one chip is better than two chips.” Their thinking is simple and direct: inter-chip communication consumes more power than on-chip communication, and two chips have significantly more transistors than a single chip with equivalent functions. However, common sense is not always correct.

As DSPs begin to integrate on-chip functions such as accelerators, dedicated communication blocks and network peripherals, their capabilities are becoming more powerful and useful to system designers. However, if the chip is left on only to perform simple routine processing or supervisory procedures, a lot of unnecessary power consumption will be generated.

The average current consumption of the system, not the instantaneous current consumption at a given time, determines the battery life. Therefore, to extend the battery life, the average current consumption must be reduced. Typical high-end processors support clock scaling and other power-saving features when the device is in operation, but if it is not in operation, it is difficult for the device to have excellent power saving performance. Many high-performance processors consume more than 50 to 100 μA in stop mode. Although this current consumption may seem acceptable at first glance, it is important to know that this is the continuous power consumption when the processor is shut down and cannot perform any tasks without external restart.

For this and other higher-end processors, maintaining a low-power state while energizing or executing a system or performing supervisory tasks can consume tens of milliamps. This means that if a system relies on a high-end processor to perform supervisory tasks, the battery life may be no more than a few days.

However, if system and supervisory functions are implemented through other devices that can manage the power supply to the main processor, the average current consumption of the system can be significantly reduced.

For some applications, replacing the DSP with an MCU to perform system supervisory tasks is a very smart design decision. Determining whether a dual-processor system architecture is the right choice depends on many factors. The application itself is the most important factor, because most designs also have to consider space and cost constraints.

For example, power monitoring, reset supervision, and power sequencing are the most basic supervisory functions required by the system. Many current SoCs have multiple power rails that must be properly sequenced at power-up for proper operation. Fixed-function devices can perform all of these functions, but they cannot meet other system requirements and cannot shut down the main processor when not needed. Replacing fixed-function devices with small, low-power microcontrollers can add the ability to manage the power supply to the main processor while also implementing sequencing, monitoring, and supervision functions.

Low-pin-count, low-power microcontrollers can achieve this function. For example, the MSP430F20x1 and MSP430F20x2 devices launched by Texas Instruments (TI) are both 14-pin microcontrollers with a comparator and 10-bit ADC respectively. They consume less than 1μA in standby current and only a few hundred microamperes when running.

Figure 1. A small microcontroller manages the power sequencing of the main processor and implements power management.

Figure 1 shows an example of a small microcontroller controlling the power sequencing of a main processor and implementing power management. A software routine on the microcontroller can start the main processor regulators in the correct order and use the internal ADC to confirm when the power rails reach the appropriate voltage. When the main processor is not needed, the shutdown feature of the regulator can be used to shut down the main processor, saving the main processor power consumption from 70μA to several milliamps.

A better understanding of the respective contributions of the DSP and MCU to power optimization is a key factor in determining when to use the two together in a design. Designers can maximize power efficiency by using two ultra-low-power processors for main processor cycles and supervisory functions, while gaining unique performance, integration and cost advantages, so that the battery life of dual low-power processor systems can exceed that of single processor systems. [page]

Power supply for DSP

DSP chip designers have adopted many low-power technology solutions, such as reducing the operating voltage and dividing the chip into multiple clock domains. Most of these solutions are executed in the background. System designers do not have to control these features too much to gain related benefits.

However, it is up to the system designer to play a full role in how the application will be executed during the DSP selection process. The following four important characteristics should be considered when selecting the best DSP.

● Use large-capacity on-chip memory: On top of the normal power consumption of the application, each time an off-chip memory call is executed, additional power is consumed. If external RAM is used, it must be powered continuously, which is a continuous power consumption process.

● Choose a DSP that provides a high degree of control over peripherals, as this directly contributes to further power reduction: Several DSPs can automatically shut down on-chip peripherals when they are not in use, or allow system designers to manually manage peripheral states. However, this feature has certain limitations in granularity.

● Choose a DSP that offers multiple standby states: The more options you have, the better the power savings will be in the long run.

● Choose a DSP that offers development software dedicated to optimizing power and minimizing power consumption: The selected tools should make it easy to scale the chip's voltage and frequency, manage power states, and measure and analyze power consumption to evaluate various design alternatives.

Power supply for MCU

The best starting point for optimizing an MCU for low-power operation is to use an ultra-low-power process to manufacture the MCU, which can reduce the leakage current of the transistor to an extremely low level. High-performance process technology will cause the DSP to consume more power, and correspondingly, a semiconductor process optimized for low power consumption may limit the peak processing performance of the MCU.

Clock speed is the most obvious limitation. For example, TI's MSP430F20xx is a low-pin-count MCU family, shown in Figure 2, that uses its unique Very Low Power Oscillator (VLO) technology to achieve standby mode currents as low as 500nA and a maximum speed of 16MHz. VLO technology enables the MSP430F20xx to fully automatically control the clock speed in ultra-low power standby mode and achieve automatic wake-up without external components, allowing systems such as smoke detectors or home thermostats to continue to work for 10 years without replacing batteries.

Figure 2 MSP430F20xx MCU structure diagram

While achieving 500nA standby power consumption, it also ensures that all device fault protection safety features are supported, such as zero-power brown-out reset (BOR) function, which can achieve both ultra-low power and extremely high reliability systems. Before the introduction of VLO, designers had to use external crystal oscillators or oscillator circuits to achieve ultra-low standby power consumption. VLO eliminates the need for external components and reduces system component count, cost, and board space, which are key requirements for portable applications. In contrast, for example, the TMS320C5506DSP has a standby power consumption of 10μA, which is 20 times the standby power consumption of the above technologies.

Using smart peripherals is also an effective IC design strategy to reduce power consumption. In the past, MCU peripherals were driven by software executed by the CPU. Although this does work efficiently, the CPU always needs to be in a working state. By designing interrupt-driven peripherals with minimal software servicing, the CPU can be kept in standby or idle mode most of the time.

In addition, the MCU selected by the system designer should also have ADC automatic input channel scanning function, hardware conversion start trigger, and DMA data transfer mechanism, etc. These features can also automate repetitive data sampling and minimize the CPU running time.

Figure 3: Two clocks are better than one for MCU power efficiency

The clock system of the MCU can also play a major role in power saving. Figure 3 shows dual clock operation from a single crystal. Two clocks are better than one for the power efficiency of the MCU.

The MCU's low-power peripherals use a low-frequency auxiliary clock (ACLK). Low-frequency, low-power operation typically uses a 32kHz external oscillator to support real-time clock functionality. A high-speed digitally controlled oscillator (DCO) can be used as the main system clock (MCLK) source used by the CPU and high-speed peripherals.

In addition to saving power by using a low-speed clock for some peripherals, TI's MSP430 MCU also integrates an ultra-low power oscillator (VLO) on chip, which can be used as the clock source for ACLK. In the standby power operation mode (LPM3), ACLK remains running and all interrupts are enabled. At this time, the typical current consumption of the MSP430 device is less than 1μA.

In addition, the DCO can enter the operating state and fully stabilize in less than 1μs, without intermediate steps. This enables "instant-on" high-performance processing functions, rather than the longer startup time required by using a second crystal or two-speed startup. This not only saves time, but also helps reduce power consumption.

Dual processor power supply

As mentioned earlier, integrated functions such as accelerators and dedicated communications and networking peripherals can cause the DSP to unnecessarily consume power when performing simple functions.

For some functions, an MCU is more suitable than a DSP, for example, keeping a real-time clock running or managing a battery charging routine. Offloading these tasks to the MCU helps better utilize the DSP's MIPS rate budget, which is very important. [page]

The advantages of a dual-processor architecture are clear. For example, if a system relies on a high-end processor to perform supervisory tasks, the battery life may be only a few days. Specifically, a typical NiMH AA battery is rated at 2500mAh. If the average current is only 1mA, it will take 119 days to drain the AA battery. If the average current is increased to 10mA, the battery will only work for 12 days before it dies.

By using a dual processor system, power optimization can be achieved through the following system or supervisory functions: maintaining real-time clock; power sequencing; power supply monitoring and reset; keypad or human machine interface management; battery management; display management.

When describing the MCU low power consumption technology, we have discussed the real-time clock function. By simply expanding the above principle, we can use the MCU to provide the clock for the DSP.

Managing DSP power

Many modern DSPs have multiple power rails that must be power-sequenced for proper operation. Typically, these rails include core rails, DDR rails, and I/O rails. Although fixed-function devices can be used to perform power sequencing, they are not scalable to support other functions.

Figure 4 DSP power sequencing implemented using MCU

As shown in Figure 4, using a small, low-power microcontroller, such as TI's MSP430 MCU, instead of fixed-function devices can implement sequencing, monitoring and supervision functions to achieve power management for the main processor (such as TI's TMS320C550x DSP).

A software routine on the MSP430 MCU enables the C550x DSP regulator circuits in the correct sequence. The MSP430 uses its internal ADC to confirm when the power rails have reached the proper voltage. If the C550x is not needed, the shutdown feature of the regulator can be used to shut down the main processor.

In fact, the MCU can control the voltage and frequency of the DSP by communicating directly with the VXO to control the voltage, or by communicating directly with the PLL to control the frequency. The obvious benefit of doing so is that the MCU can put the DSP into an effective standby mode after the DSP completes its computationally intensive tasks.

The monitoring process is bidirectional, in other words, the MCU can query the DSP to see how busy it is. In this mode, it can function as an intelligent controller. On the other hand, the DSP's internal monitoring function can also function. Since the MCU can be read and written, the DSP can inform the MCU to speed up or slow down the clock according to the application needs.

Human-machine interface management

We can reduce the power consumption of the DSP by interacting with the MCU and avoid using the DSP internally to perform simple tasks. However, the greater benefit of this approach is that system designers can offload some of the supervisory tasks that are performed by the DSP in a single-processor system to the MCU.

The MCU can easily meet the requirements of keypad operation, and the power consumption is much lower than that of the DSP. For example, the standby current of only 500nA is much lower than that of the DSP. By ensuring that we interrupt the DSP only after the key is pressed and released, the MCU can avoid the current consumption caused by the key being stuck, which is not uncommon in some handheld devices.

In addition, the 16MHz MCU can also support the management of the user display. To further improve energy efficiency, the MCU should have an integrated segment LCD driver to handle four multiplexed data streams.

In addition, the MCU should have integrated functions to communicate with peripherals via standard SPI, UART, I2C or RF, and should be able to automatically boot from low-power modes without polling.

Battery charging and charge management is another supervisory function that is typically performed with fixed-function devices, but can also be performed by a microcontroller. The microcontroller's ADC can be used to measure voltage; timers and software can be used to manage charging; and pulse-width modulated outputs can be used to provide charging waveforms. Microcontroller vendors' application reports often describe how their products accomplish these functions and also provide code samples. For example, application report number SLAA287, titled "Li-Ion Battery Charger Solution Using MSP430," which can be downloaded from the TI website, describes a solution for the MSP430 product.

Once the system is grouped, it is necessary to connect the MCU and DSP together to ensure that they can communicate with each other. Most of the data required for the MCU to interact with peripherals must be shared with the DSP. Of course, this will place certain requirements on the technical specifications of the MCU. We should look for products with at least 16 GPIO ports to support communication between processors. In addition, the selected product should also support on-chip SPI and I2C interfaces, and the battery charging function requires a 10-bit ADC.

In most systems, we can build a simple protocol that allows the DSP to issue basic commands to the microcontroller. In some cases, if the main processor and microcontroller are provided by the same manufacturer, the two can be well integrated through application reports and related solutions.

in conclusion

DSPs have made significant progress in reducing power consumption, but the high-performance process technologies used to manufacture DSPs have limited their potential. When chips are in standby or idle mode, they are particularly susceptible to power consumption caused by transistor leakage current. Ultimately, designers need to determine whether their application should use one processor or a combination of two processors based on calculations, measurement analysis, and trade-offs between the DSP and/or MCU.