Low power MCU design: chip and system-level considerations - Embedded.com

Low power MCU design: chip and system-level considerations

Reducing power consumption has a major impact on every aspect of our lives. At a macro level, the benefits have been well-documented: lower electric bills for consumers, reduced load on utilities, and fewer batteries in landfills. In short, saving power is good for both the environment and the pocketbook.

Due to the growing use of electronics worldwide, reducing power consumption must begin at the microchip level. Power-saving techniques that engineers have designed in at the chip level have a far-reaching impact, especially when involving microcontrollers that serve as the engines behind most of these electronic devices.

From a system-design perspective, identifying which microcontrollers are truly low-power requires designers to navigate through the myriad claims of various semiconductor vendors. Because of the varying and confusing metrics vendors use, this objective is a complicated task.

This article briefly describes the main factors that you need to consider when analyzing competitive microcontroller alternatives. At a basic level, you can define microcontroller power consumption as the sum of active-mode power and standby, or sleep-mode, power. However, another important metric to keep in mind is the amount of time it takes for a microcontroller to move from a standby state to an active state.

Because the microcontroller cannot do any useful processing until all of its digital and analog components are fully settled and operational, adding in this wasted power is important when calculating total power consumption. Thus, the device's total power consumption is the sum of its active-mode power, standby power, and wake-up power.

Because every application is different, system designers have a tendency to give more weight to some of these elements than others. For example, some applications, such as water meters, spend most of their time in a standby state, so their long duty cycles require low standby-power consumption.

Other applications, such as data loggers, often enter and exit active states, so it is critical to limit the time they spend in wake-up-transition modes (Figure 1). However, a vendor developing a compelling microcontroller cannot attempt to guess which of these metrics is most important but instead designs a system from the ground up that focuses on minimizing any power consumption.

Accomplishing this objective requires strong mixed-signal expertise to address both the architectural-level and the circuit-level challenges necessary to minimize power in both the analog and the digital domains.

Active-mode current

For a CMOS logic gate, the following equation yields dynamic power consumption: C×V2×f, where C is the load capacitance, V is the supply voltage, and f is the switching frequency. The capacitance term is a function of the design and processing technology, and the frequency term is a function of the application's processing requirements.

However, as you can see from the equation, the supply voltage has a disproportionate impact on the overall power that the microcontroller consumes. Therefore, adding voltage regulation to the microcontroller design can yield significant active-mode-power savings by providing a lower, steady supply voltage to the microcontroller's circuitry.

Switching-type converters may be a possible approach for this application, but they better suit regulator environments requiring large voltage-conversion ratios.

However, in battery-type applications in which the average voltage-conversion ratio approaches 1-to-1 at the end of the battery's life, a better approach would be to add an on-chip low-dropout linear voltage regulator because it offers acceptable efficiency with less complexity and lower cost than a switching approach.

To illustrate the benefits of using a low-dropout regulator, it is helpful to restate the CMOS dynamic power equation as C×V2×f=V×(C×V×f)=V×I, where the dynamic current, I, equals C×V×f. It is common to normalize the dynamic current to a frequency of 1 MHz and a particular supply voltage.

For example, one recently introduced ultra-low-power microcontroller has a dynamic current consumption of 160 μA/MHz at 1.8V. Without supply regulation, this metric would increase to 160×(3.2/1.8)=284 μA/MHz when the supply voltage is 3.2V. With a low-dropout regulator, the battery current remains at 160 μA/MHz across the entire supply range (Figure 2).

You can use this advanced power architecture to maintain a constant active current over the full operating-voltage range, and it can help you achieve a significant savings in power consumption. Therefore, it is important to determine the microcontroller's current consumption when operating across the entire operating-voltage range, not just at the 1.8V minimum operating condition that microcontroller vendors commonly quote. Quoting an optimistic current number that assumes anything less than a typical voltage supply does not accurately reflect how applications find use in the real world.

As an example, in systems requiring two AA or AAA coin-cell batteries, the batteries most often operate near their initial 3V voltages. Therefore, the quoted 1.8V specification can be deceiving because, from this perspective, most microcontrollers consume approximately 50% more power than the amount that their vendors commonly quote.

Furthermore, because power consumption is directly proportional to switching frequency, system designers should normalize the quoted current numbers down to a current-per-megahertz basis. By combining these two factors, you can perform a side-by-side comparison of microcontrollers using current consumption per megahertz at 3V.

Some vendors attempt to confuse the issue by equating megahertz to system-clock speed when the value that is truly meaningful is instruction-clock speed. This substitution is deceiving because system clocks can operate at least twice as fast as they can execute instructions, thereby at least doubling their effective power consumption.

It is therefore important to normalize specifications to instruction-clock speed. By doing so and by using a typical supply voltage, you can properly derive the actual active-mode current-consumption budget.

Standby current

Achieving maximum energy efficiency and, therefore, battery life requires ensuring that each microcontroller task consumes the minimum possible current at the minimum possible voltage for the shortest possible duration, so that the device spends most of its time in a low-power sleep mode.

In some applications, the sleep-mode current is the parameter most responsible for overall energy consumption. However, engineers often overlook the fact that leakage current is the primary limitation on the absolute minimum sleep current a microcontroller can achieve. For example, a 20-input device that has an input-leakage-current specification of 100 nA could consume as much as 2 μA of power during sleep mode.

A number of factors, the most important of which is process technology, affect leakage current. In some cases, vendors use a 0.25- or 0.35-micron process technology to reduce the leakage-induced sleep current, but this choice comes at the expense of a higher active current.

In other cases, microcontroller vendors use 0.18-micron or smaller process technologies to reduce active-mode current, but this choice comes at the expense of higher leakage currents. One approach to this problem is to apply mixed-signal expertise to implement an advanced power-management unit that limits leakage and enables ultralow sleep current regardless of the underlying process technology in use.

When using process technologies of 0.25 micron or smaller, minimizing sleep-mode current requires cutting power to the digital core. Modules that operate in sleep mode, such as power-management circuits, I/O-pad cells, and real-time clocks, must operate from the unregulated voltage supply to avoid burning additional current in a low-dropout regulator.

Cutting power to the digital-core logic also prevents its off-state leakage from contributing to the sleep-mode current; however, the microcontroller must preserve RAM contents and the state of all registers during sleep mode so that execution can resume where it left off.

You can perform this preservation using a low-current sleep-mode latch-bias scheme or retention latches that can hold the state in sleep mode without significant leakage. The microcontroller also requires some form of continuous supply-voltage monitoring, such as brownout detection, to reset the device in the event that the supply voltage drops below the minimum retention voltage, which could corrupt the operating state.

It is therefore important to examine the underlying leakage-current specifications to determine which microcontroller vendors have applied their mixed-signal expertise toward solving this complex problem.

Designers should also consider the fact that most vendors offer many standby-current options. Most suppliers highlight their absolute lowest sleep-mode current, which often corresponds to the current the system consumes when the real-time clock and brownout detector are disabled.

Some vendors go a step further and quote a shutdown-mode current that does not retain memory and requires a reset to wake up, which in general is not a practical mode.

Therefore, because most applications require full RAM and register retention, it is important to perform side-by-side comparisons using the standby/sleep-mode current with real-time clock and brownout disabled with RAM retention, standby/sleep-mode current with real-time clock disabled and brownout enabled, and standby/sleep-mode current with real-time clock and brownout enabled. You can then use the correct values when calculating the overall standby-mode power budget based on the duty cycle of the application.

Wake-up energy

Systems that use sleep modes can waste a significant amount of power waking up the microcontroller and preparing it to acquire or process data. In some applications, a microcontroller can often use just as much energy when coming out of standby as when the device is fully processing data.

Therefore, it is important to design a microcontroller to wake up and settle in a short time to minimize the amount of time it spends in an energy-wasting state. The microcontroller should be able to exit sleep mode from either an external trigger event or an internal timer.

The most flexible periodic-wake-up source is a real-time clock that can operate from an external crystal oscillator for applications requiring accurate timing or from a low-frequency internal oscillator that eliminates the need for a crystal in applications not requiring high accuracy. Avoid using a slow-starting crystal oscillator for the high-speed system clock; an accurate, quick-starting, on-chip oscillator is a better alternative.

Because many products periodically wake up to sample an input using on-chip ADCs, it is also important to allow enough time for both the digital circuitry to wake up and the analog circuitry to settle before beginning to make valid measurements. The start-up behavior of the analog modules can have a major impact on the amount of time the system spends in active mode; voltage regulators or references using external decoupling capacitors can take many milliseconds to settle.

Microcontroller vendors sometimes quote only the wake-up times for the digital circuitry and ignore the time it takes for the analog circuitry to settle. Therefore, it is important to analyze the overall wake-up and settling time for both the digital and the analog circuitry to factor in the true cost of this wasted energy.

Other considerations

Other ways also exist to further reduce power in a system. For example, you can use two AA or AAA batteries in some configurations because microcontrollers often can typically operate with voltages as low as 1.8V with fewer functions. Or they may lack an ADC or have a slower instruction clock. An innovative way to reduce power and environmental impact is to convert the design to a single-battery configuration in which the battery operates at 0.9V to the end of its useful life.

To enable this approach, a microcontroller must integrate a highly optimized dc/dc converter that can operate to the lowest usable voltage of the battery, which, in the case of alkaline chemistry, is 0.9V.
This approach can also save the supplier, the consumer, or both the cost of a battery.

Another method of reducing power is to use highly integrated microcontrollers that include ADCs, DACs, and other peripherals because the microcontroller can have control over enabling and disabling these peripherals when the applications require them. For example, some microcontrollers offer specialized low-power ADCs with burst modes that can take analog measurements when the CPU is off in an effort to further minimize power consumption.

Mike Salas is vice president and general manager of Silicon Labs' microcontroller products.

This article was previously published on Embedded.com’s sister publication, EDN Magazine. 

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.