Choosing the right low power processor for your embedded design -

Choosing the right low power processor for your embedded design

In early days, getting a lower power CPU typically meant sacrificingfunctionality, running at reduced clock speeds, or waiting for new lowpower process technologies to reduce both Standby andActive power. This is nolonger the case by any means, and the processor landscape has changeddramatically.

Advances in processing technology along with innovative chip designsand high-granulation power management software have brought entirelynew families of low power processors where designers no longer need tomake sacrifices in their system designs.

Of course, no one device “has it all,” so engineers must considertheir system requirements carefully and then examine the now expandingrange of low power processors to see which one best fits theirapplication requirements.

This article summarizes the state of the art with aproduct-selection matrix (Table 1, below). One axis shows the followingdesign criteria that are of chief concern to system designers:

* Power
* Performance
* Integration
* Time to Market
* Price

The other axis lists the major processor variants based on theirfeature sets. This article then explains the meaning behind the genericcriteria and how various types of processors earn their rankings in thetable.

This information achieves two goals: first, it alerts systemdesigners to the newest types of processors on the market, some ofwhich are relatively new and about which they might not be familiar;second, given this ever larger product palette, it helps them narrowdown the selection of the best chip for a given design.

Examine the criteria
To help you sort through the various low power devices, refer to Table1, which grades the major types of low power processors according toseveral criteria of interest to designers. The first thing to note isthat these criteria are all closely interrelated.

For instance, integrating a large number of functions on a chip suchas multiple cores, analog features, large memories or many peripheralscan reduce overall system power, cost and time to market. Extensiveintegration of this nature, though, can add unwanted power consumptionand make programming more complicated, thus extending time to market.

Criterion #1: Power
For many of today's designs, this is the single most importantcriterion. In portable products, extended battery life is a bigconsumer plus. In many infrastructure applications, lower powertranslates to less heat dissipation ” and heat-dissipation “envelopes”can be the limiting factor for channel density or feature additions.

There are also designs with a power budget such as USB-operatedproducts or automotive aftermarket products running from a car batterythat are allocated only a certain milliwatt budget for operation.

Power should be more properly viewed from a systems perspective. Theright mix of peripherals on a chip results in more overall system powersavings not only because off-chip devices consume extra power, but alsobecause it takes a lot more power to move data across traces on a PCboard than it does to move data within a device itself.

For individual devices, energy efficiency starts with the inherentbenefits of process technology, but this is only the beginning of whatadvanced processors offer in this regard. Power consumption can bebroken into two main modes: first, active power consumption, which isperformed with transistor switching and takes place during ongoing dataprocessing; second, static power consumption, which takes place wheneither limited or no data processing takes place and various componentsgo into a type of Sleep mode.

Several techniques are used within active power management:

* Dynamicvoltage and frequency scaling (DVFS). Here, clock rates andvoltages are lowered by software command depending on the performancerequired by the application scenario.

For instance, even though the ARM on a multimedia processor might beable to run at 600 MHz, all that power is not required in every case.Instead, software can select from predefined operating performancepoints that run the processor at specific rates.

* Adaptivevoltage scaling (AVS). This is based on the fact that siliconmanufacturing yields parts with a distribution of performancecapabilities; for a given frequency requirement, some devices (known as”hot” devices) can achieve that level of performance with a lowervoltage than can “cold” devices.

In this situation, a processor senses its own performance level andadjusts voltage supplies accordingly to compensate for variations inprocessing, temperature and silicon degradation.

* Dynamic powerswitching (DPS). This determines when a section of a device hascompleted its current tasks, is not needed at the moment, and puts itinto a low power state. An example of this granulated power control iswhen a processor enters a low power state while waiting for a DMAtransfer to complete.

Static power management takes place when either limited or no dataprocessing occurs, selected components can drop into a very low powermode, and where the system waits for a wakeup event.

Handled by a technique known as static leakage management, it canresult in several low power modes from Standby to full Power Off. Thechoice of which low power static mode is chosen depending on whatdegree of memory retention and/or a fast wakeup time is needed.

Thanks to these features, most low power processors spec a standbypower in the range of 15 mW and a peak operational power below 400 mW.However, some fixed-pointdigital signal processors drop those figures to 0.50 mW standby and75 mW peak even though it contains a FFT coprocessor, as much as 320kbytes of memory and I/O peripherals.

In the table, most of the devices implement many if not all of thesepower-saving features and get an “excellent” rating. The ones listed as”good” are the highest performing chips, generally with multiple cores,which naturally draw somewhat more power.

Criterion #2: Performance
This criterion is important because added processing capability oftendifferentiates end-user products by enabling new functions as well asmore channels per cost or area, faster datarates as well as denser andhigher-quality compression schemes.

In looking at performance, engineers should look beyond MHz and alsoconsider parallelism. A big performance boosts come from chips thatintegrate a DSP, ARM or coprocessor in various combinations, an examplebeing the OMAP platform. Engineers canthus partition their code to run on the most appropriate core.

Parallelism is a benefit you can get even on a device with just onecore. For instance, a single CPU in low power fixed-point DSPsĀ have extremely high processing capabilities thanks to eight instructionunits that run in parallel at 300 MHz. For the same low power budget,the device offers 2X the processing capabilities as other low powerprocessors on the market.

Besides integrating processing elements, integrating other systemcomponents can lead to considerable performance improvements. Forinstance, having sufficient on-chip memory means that a CPU can runcode much faster than if it had to import and export data morefrequently.

No matter what kind of system in development ” whether multimediaappliances or those that have limited functionality but need the lowestpossible power ” designers can choose a processor with exactly theamount of processing power they need.

In the table, the range of performance going from “fair” to”excellent” is generally a function of how many cores and on-chipperipherals a given device has. As always, one tradeoff forperformance, though, is typically power consumption.

Criterion #3: Integration
Clearly this aspect is closely related to performance. As just noted,certain chips offer designers the choice of either or all of thefollowing on the same chip: DSP, ARM9 or a coprocessor.

With regard to integration, though, other essential systemcomponents can fit on today's devices. A good example is integratedmemory, which lowers total system price, saves system power and easesdevelopment. Some low power processors incorporate almost half amegabyte of memory directly on the chip, such as in the OMAP-L1xapplications processors, and in many cases this eliminates the need forany external memory.

Today's processors, however, can integrate a much wider range ofperipherals, including analog components. A prime example is an SAR(successive-approximation register) A/D converter. SARs are useful, forinstance, for interfacing to touch-screen displays common in consumerappliances.

Another example is a uPP (universal parallel port), which allowsdirect connection to a wide range of other parts on a system such ashigh-speed ADCs or FPGAs. On today's low power processors you can alsolook for on-chip support for networking with Ethernet MACs, USB 2.0,Serial ATA (SATA) for mass storage, SDIO for I/O functions like WLANsupport, LCD controllers, and video-port interfaces.

In the table, a rating of “excellent” refers to devices that havemultiple cores or a coprocessor as well as a variety of peripherals;the “good” rating applies to devices with a single core but largeamounts of memory and peripherals; a “fair” rating goes to devices thathave fewer peripherals but that are power stingy and less expensive.

Criterion #4: Time to Market
This aspect is taking on ever greater importance as the rate ofinnovation in consumer products continues to increase and product lifecycles are shrinking from years to months. No sooner is your latest,greatest product on the store shelves when a competitor brings outsomething a few months (or weeks) later with significant new featuresthat draw consumers' attention.

Time to market is related closely to the level of integration.Obviously, if components are on-chip, engineers require lessdevelopment and debugging time because there is no need to develop theinterfaces and data-exchange facilities necessary to coordinate theactivities of multiple chips. There is also less effort required indealing with pc-board interconnects and working with separate drivers.

When many cores or peripherals are integrated on a chip, however,engineers need the proper software tools to help them manipulate thecomponents. For instance, with a combination of an ARM and a DSP, agood toolset will allow the development of applications that need theresources of both cores but within a single programming environment.

In addition, engineers should also look to see what other tools theprocessor vendor offers in terms of third party algorithm librariesoptimized for the various cores, support for third party tools such asSimulink from Matlab or LabVIEW from National Instruments,evaluation/development boards and even a variety of operating systems,even open source options. All of these factors are important inreducing development time and getting products to market on or beforedeadline.

A final aspect not to be forgotten is that floating-point DSP devicesĀ are less complex to program. In many cases, developers can write theircode on desktop PCs using familiar tools such as Simulink and LabVIEWand then port the code to the DSP with very little modificationsneeded, if any.

In general, though, it is safe to say that the higher theperformance on a given chip, the longer the development time. For themore-complex products that require this level of performance itobviously takes a longer time to develop and debug the code.

Finally, engineers should always be thinking ahead to the nextgeneration of their products. In some markets, standards are fluid, butcompanies want to get an early jump into the market. Thus, designersmust build “future proof” products that can be upgraded to reflectchanges in standards or add new features.

It's thus important to look at a family of processors and examineits intercompatibility both in software and pin-for-pin compatibility “if I need more computational power, can I later add it with only veryminimal changes to the overall system design and code?

In the table, a rating of “excellent” applies to devices with broadsupport in both hardware and software. A “good” rating goes to deviceswith a lower level of integration, meaning a few more off-chipperipherals or memory and the associated design effort.

Criterion #5: Price
When evaluating this criterion, engineers should look beyond chipprices, which themselves are dropping such that most low powerprocessor now generally cost below $15, and depending on devicefeatures prices can drop to levels even as low as $4.00.

While the cost of each component is critical in consumerapplications, it plays less of a role in infrastructure or commercialapplications where the cost of ownership and efficiency tend to commandmore attention.

Engineers should rather consider total system cost. For example,returning to memory, if you can run all of a product's algorithms fromon-chip memory, you've saved a dollar or two just for those extramemory chips that are no longer needed.

Significant system savings (up to $9.00) can be saved on integrationcombinations such as SATA, Ethernet, memory, USB 2.0, the ARM9 seen inthe OMAP-L1X applications processors and other highly integratedperipherals mentioned in the Integration section.

Besides the price of chips, engineers should also evaluate ease ofdevelopment, an aspect that encompasses software and hardwaredevelopment tools, technical support, training, third party support,documentation, engineering time/overhead and NRE development expenses.The bottom line is that faster development can lead to higher qualityend products because valuable time and money are spent ondifferentiation rather than building the design infrastructure.

Thus, engineers should also consider not only the price ofdevelopment boards and emulators but also their quality and how muchthey can speed development projects. High-quality IDEs and compilersgive designers more visibility into their design and reduce time tomarket.

Look for silicon vendors who offer royalty-free operating systems,off-the-shelf verified code from third parties such as the codecs usedin DSP-based designs as well as frameworks that allow designers to getgoing quickly on their designs.

In addition, don't forget the cost to layout and manufacture theboard. Not only is the number of devices important, so is the pitch ofdevices ” small-pitched devices are more expensive to lay out andmanufacture at the system level.

In the table, price is generally inversely proportional to thenumber of cores and on-chip peripherals. The more such components,obviously the more expensive the device and the design effort becausethese are targeted at the most sophisticated portable systems. Forexample, the only category that gets a “fair” rating is thehigh-performance application processor that can have a DSP, an ARM anda coprocessor.

Low power applications
Even with the help of this table, it's not easy to select the bestdevice for a given application. There will always be design tradeoffs.But a brief discussion about application requirements might providesome guidance. Applications that require low power consumption havegreatly expanded, and it helps to categorize the major areas:

* Products that are plugged in or USB powered such as a hands-freecar kit, GPS dongle, touch screen or a speakerphone

* Applications where consumers expect batteries to last at least afull day such as a wireless microphone, musical instruments,noise-reduction headphones, wireless printers, and even multiparameterportable medical instruments

* Applications that should allow a battery lifetime of up to twoweeks, such as a music recorder, e-book, door-lock fingerprintauthorization or single-parameter portable medical instruments

Another way to categorize applications is by separating them intogroups based on functionality. One concern is high precision in aportable device, such as in a musical instrument or audio product thatrequires a high dynamic range. This level of precision and dynamicrange typically requires a floating-point processor , whichhas power consumption from 15 mW.

Now consider applications that rely on a feature-rich GUI. Here, adevice that offers ARM-based processing is a good choice. Thanks to theARM + DSP integration on devices such as the OMAP-L1x applicationsprocessors, there is plenty of capacity to run the GUI as well ashandle sophisticated processing tasks.

Then there are products where consumers demand long battery life ina portable device, among them being portable audio recorders/ players,e-books, portable microphones or even home medical monitors that fit onthe wrist. Processors that focus on low standby modes can enable weeksof battery life through high utilization of deep sleep (6.8 microwatts)and standby states (0.5 mW).

As stated often in this article, all of the selection parameters for alow power processor are closely interrelated. It's always been a caseof the highest performance implies the highest power consumption “except today power levels have dropped across the board to where youcan find a low power processor for virtually every need.

John Dixon is the Low PowerProcessors Product Line Manager at Texas Instruments, Inc.. In thisrole, Dixon manages the marketing and strategic direction of TI's broadmarket low power processor platforms including the TMS320C550x, TMS320C674x and TMS320C640x DSPs. He has sixyears of marketing experience with TI's Semiconductor Group. Previousto his work at TI, Dixon balanced both business and technical roles inan international management consultant position and multiple technicalelectronic research positions.

Leave a Reply

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