Modern electronics have come a long way since the vacuum tube. The first computers occupied entire floors of buildings, radios were large systems integrated into cabinets, and the closest thing to a portable device was a two-way radio system that was worn on the back. Now, computers far more powerful than those that landed man on the moon are produced at dollars apiece, can be integrated into almost any everyday device, and can connect to the internet.
But no matter the achievements made by engineers, devices are never too small, never too efficient, and never too portable. Simply put, no matter how much we try, we always want to do better, and this was famously demonstrated by Steve Jobs with the development of the iPod.
During the early developmental stages of the first iPod, Steve Jobs asked if the engineers could reduce its size. They said that it could not be reduced any further, as they had used all the space possible, but Steve Jobs remained unconvinced. To prove his point, he submerged the prototype into a fish tank, only to show air bubbles leaving the iPod. If there was air inside the iPod, then there was space not being used (or in this case, too much space).
Power problem - to decrease power usage or increase capacity?
Creating portable devices requires the device to carry its own form of power. Simple devices such as calculators can rely on small solar cells, as the calculator uses extremely small amounts of power, but devices such as smartphones, laptops, and smartwatches cannot use such power sources.
Instead, they require the use of internal batteries (generally lithium-ion), and this leads engineers into an energy-balancing issue. If the device uses too much power too quickly, the resulting device will have a short operating time, meaning it will require more frequent charging. To extend the operating time of the device, a designer has two options: Increase the battery size or decrease the power consumption.
Increasing the battery size is a very fast and simple option, but it comes with some big disadvantages. The first major drawback is the increased size of the device. As portable devices are designed to be as small as possible, increasing the size of the battery can significantly add to its size. The second major drawback is the increased weight of the device. A battery’s weight is generally proportional to its power output, and as such, doubling the stored energy doubles the weight of the battery. The third major drawback is price. Doubling the battery doubles its cost, and batteries such as those based on lithium-ion technologies are not cheap.
Decreasing the power consumption of the device is an effective option for extending the operating time, but it does come with its own challenges. The quickest power-reduction option is usually to decrease the brightness of screens and slow down the system clock of the main processor. However, this will result in a dimmer display that may be harder to see, and a slower system clock will result in lower system performance. Decreasing power can also be done by turning off unused peripherals as well as power-cycling peripherals such as wireless systems, but this can increase network latency as wireless systems power up and reconnect between each data transmission.
However, the use of alternative technology can sometimes provide designers with the ideal option for increasing battery life. The replacement of older technology with newer technologies can allow for faster CPUs, better displays, and better energy-saving states that provide energy improvements while not impacting system performance.
Custom systems-on-chip (Apple M1)
A system-on-chip (SoC) is a single semiconductor device that integrates all system components onto a single chip (minus memory and I/O interfaces). SoCs have become extremely popular thanks to their small profile and low cost, but while their processing capability generally pales in comparison with mainstream processors (such as those provided by Intel and AMD), their energy consumption is a fraction of those processors. For example, the Raspberry Pi, which is powered by a Broadcom SoC, consumes less than 5 W and can provide users with a processing platform capable of most everyday tasks, including email, web browsing, and video.
While off-the-shelf SoCs have provided designers with a low-power computing platform, those SoCs are still designed for a wide range of users and, as such, will integrate hardware that may go unused in many designs. If that hardware cannot be deactivated, then energy is potentially being wasted in said unused hardware. Furthermore, the inclusion of unused hardware results in a silicon die that is not maximizing its capabilities for the intended application.
A new trend is forming in the industry whereby product designers are creating their own SoCs instead of using off-the-shelf systems. One example is the Apple M1, which is being used to power the next generation of Apple MacBooks and integrates eight ARM 64-bit cores (four high-power and four high-efficiency), neural-network hardware, and an integrated GPU. Overall, the M1 has 1,024 ALUs and a maximum floating-point performance of 2.6 TFLOPs. According to Apple, the M1 has the best performance per watt, and this is enabled through the use of a totally custom design.
Apple is not the only company looking into custom silicon; Google and Amazon are also developing custom hardware to power their data centers to provide more energy-efficient computing while maximizing the computational performance of silicon.
While the development of custom SoCs and ASICs is very expensive (and thus out of reach for most engineers), a new technology may soon change this. Instead of providing all-silicon chips in packages (such as MFL and TQFP), designers could choose various semiconductor products as bare chips, and these can then be combined together into a custom package that uses high-speed interconnects to connect the various chips together. While each semiconductor device is an off-the-shelf part, this would allow for smaller semiconductor devices that can be tailored to their intended product at a much-reduced price.
Dynamic frequency scaling (Intel hardware P-states)
Intel hardware P-states are specific to Intel CPU products, but the concept can easily be expanded into microcontrollers and other devices. Simply put, hardware P-states are initiated by the running software to put the processor into a lower-power mode. This can include reducing the clock frequency or voltage to the CPU, and such hardware states usually interface with the power supply to the system to account for this. Power reduction can also be done by dynamically powering down peripherals that are either not used at all or doing so in cycles such that they are consuming power only when needed.
However, a word of caution when developing such mechanisms: While there is nothing wrong with using a software-monitoring system to adjust the power and clock frequency of a processor, the processor (and surrounding hardware) must not do this by itself. This is because there is a protected patent (from VLSI Holdings) that describes such a system, and Intel is currently fighting a court case against VLSI Holdings over Intel’s Speed Shift technology.
Low-energy displays (Kindle paper display)
Sometimes, reducing the power consumption of a device can be done in more unusual ways, and the Amazon Kindle achieved this with the use of the e-paper display. E-paper displays are a special display technology that utilizes small, white spheres half-coated in black ink. The use of an electric field can orient the spheres, but once oriented, the electric field can be deactivated. With each pixel having one or more spheres, once an image is sent to the display, the image remains on the display even when powered down (of course, over time, it will deteriorate).
As such, the Kindle’s use of an e-paper display provides a battery operating time of weeks instead of days, and the device still has wireless and browsing capabilities. Furthermore, e-paper displays create a display that has a similar feel to ink on paper, and the lack of a backlight makes the display more comfortable to view. The lack of a backlight also significantly reduces power consumption but at the cost that the device can be used only when external light is available. The low refresh rate of e-paper displays coupled with the lack of a backlight makes them impractical for smartphones, but other devices such as smartwatches and interfaces could benefit from them.
Hardware data processing (AI hardware)
It is very easy for software and hardware engineers to move all data processing into the CPU. The lack of additional hardware creates a simpler design, and the use of a single location for execution allows the software to operate in one single domain.
Hardware acceleration, however, shifts unique data operations into dedicated hardware circuitry. Such circuitry almost always outperforms CPUs on both operations per second and consumed energy. For example, running AI inferences on a CPU is a complex procedure due to the use of vectored matrices, but GPUs (which are specially tailored for such operations) can perform the same task far more efficiently. This is why many AI systems utilize GPUs instead of CPUs.
But it’s not just AI; many complex tasks can be moved to dedicated hardware circuitry, including cryptography and error correction. As such, designers should look at what operations their design needs to complete and if those operations can be moved from the CPU to a dedicated processor. In many cases, microcontrollers can be found to integrate additional cores and hardware for such tasks.
Cloud computing (remote processing)
Reducing system clock frequencies as well as reducing the number of cores used on a device can provide major energy savings, but this comes at the cost of reduced system performance. While many system tasks such as GUI interfaces and data storage use little system resources, heavy processing such as AI inference can dramatically increase energy usage.
As such, one common method to increase battery performance of devices is to offload as much data processing as possible to a remote data center in the form of cloud computing. Data that needs to be processed is sent to a remote computer, which can perform energy-intensive calculations, and the results are returned via the internet for the device.
This method is widely used in IoT devices such as the Amazon Alexa, which uses local hardware to determine if it is being activated (“Alexa” command) but then streams spoken requests to an Amazon data center (“Play some music”). However, the use of remote processing introduces privacy concerns and should be discouraged if the data being processed is in any way private to the user (i.e., images, spoken sentences, and personal data).
Conclusion
Reducing energy consumption is very much like accounting; at the end of the day, the books need to be balanced, and it is amazing where you can cut costs or, in this case, energy. The use of alternative technologies such as e-paper can help to create unique devices, while deploying smart-energy–monitoring systems can intelligently reduce power consumption.
However, it is most likely that the biggest energy savings for devices will come in the form of chiplets, and hopefully, engineers in the future will be able to customize their own systems from the ground up using off-the-shelf chip dies in a near-identical manner to custom PCBs.
