EDNE December 2012 - (Page 30)

MANY CONTEXTS MICROPROCESSORS COPROCESSORS WORKLOAD/ ENERGY WORKLOAD/ TIME MICROCONTROLLERS MICROCONTROLLERS DSCs DSPs FPGAs FEW CONTEXTS LESS WORKLOAD ENERGY COST/ WORKLOAD MORE WORKLOAD Figure 1 This taxonomy mapping identifies the processing sweet-spot spectrum of mainstream processing architectures. The labels near the edges of the chart identify the key tradeoffs for extreme processing innovations for processors near those edges (courtesy Embedded Insights). MANY CONTEXTS 32/64+ BIT WORKLOAD/ ENERGY 16-BIT MICRO8-BIT CONTROLLERS 32-BIT WORKLOAD/ TIME FEW CONTEXTS 4BIT LESS WORKLOAD ENERGY COST/ WORKLOAD MORE WORKLOAD Figure 2 Processor bit widths are superimposed over the sweetspot spectrum to indicate the processing sizes implemented in each sweet spot (courtesy Embedded Insights). silicon area. In fact, 32-bit microcontrollers broke the $1 barrier years ago, and the smallest devices have even broken the 50-cent price point, placing those devices squarely in the price range of 8-bit microcontrollers. There are a few costs, however, that the 32-bit microcontrollers must cover but the 8-bit devices can avoid. We’ve already mentioned the depreciation on the fabrication facilities. Further, because we are assuming the 32-bit device is an ARM microcontroller, the price must cover the royalty fee due for using ARM’s IP (intellectual property), further cutting into relative margins. Also cutting into relative margins is the fact that 32-bit devices are more support intensive; this is where using a 32-bit IP enables a semiconductor company to leverage some of the support costs through shared development resources with other companies. The support costs for 8-bit devices can be lower because the target applications are generally simple in scope and size, operate at “slow” clock rates, and are supported by a fiercely dedicated and cooperative developer/ user community external to the supplying vendor. In short, there are probably multiple vectors that let an 8-bit vendor manoeuvre on price and manufacturing when a 32-bit device poses a real threat solely because of price parity. What about when the 32-bit processor meets or exceeds the 8-bit microcontroller’s energy performance? Here, the 32-bit device leverages a double-prong approach to challenge 8-bit devices: code density and the time to perform a wake/sleep cycle. Rob Cosaro, senior director for architectures and systems at NXP Semiconductors’ microcontroller business, says the company’s benchmarking research has shown up to a 50% reduction in code size when performing the same algorithm on a Cortex-M0-class processor as on an 8051. But benchmarks are tricky unless they reflect the code your design is using. For example, the Coremark benchmark from EEMBC (the Embedded Microprocessor Benchmark Consortium) contains functions that test 8-, 16-, and 32-bit CPUs, but functions such as a double link list and matrix manipulation are likely tasks you would not consider performing on an 8-bit device. The opportunity for a 32- or even a 16-bit processor to provide better code density arises when an 8-bit processor is used outside its ideal region of use, such as for operating on data larger than 8 bits (because it requires multiple data accesses to operate on a single datum), operating on data sets that exceed 16 to 64 kbytes of address space, operating at high clock rates (higher than 20 to 50 MHz), or even supporting heavy network communication stacks. In such cases, the application may be mismatched with an 8-bit processor because maintenance-related feature growth has crept into the system over a period of years. In embedded designs that are energy sensitive, the microcontroller spends most of the time in a low-power sleep mode and wakes up periodically to perform its tasks. As in the case of code density, if the task that the 8-bit microcontroller is performing is mismatched, a 32-bit microcontroller may be able to wake up, perform the tasks, and go back to sleep fast enough that it actually consumes less energy than the 8-bit device performing the same task. A key advantage of an 8-bit microcontroller over a 32-bit processor that may be able to take over its tasks is that the 8-bit device may have enabled the task to be performed at a cost- and energy-effective level several years before the 32-bit device was able to replace the 8-bit controller. The excitement in the small-processor segment centers on the smallest ones—the ones that push the cost and energyperformance limits of what has been possible. What we describe as low power is a constantly shifting target. Smaller data widths will always significantly lead wider widths in when they become able to support those small tasks. There is one other factor that will not show up in a data sheet but that matters when choosing between 8- and 32-bit devices that have achieved price and energy-performance parity: domain expertise. Though programming an 8-bit device may require fluency in assembly or even C language, 30 EDN EUROPE | december 2012 www.edn-europe.com http://www.edn-europe.com

Table of Contents for the Digital Edition of EDNE December 2012

Cover
Contents
Texas Instruments Europe
Microchip
Digi-Key
Masthead
EDN comment
Pulse
Digi-Key
Baker’s Best
Embedded World 2013
Test & Measurement World
Rohde & Schwarz
Digi-Key
Squeezing the most from battery cells with a switched-mode pump
Signal integrity
Processor architectures : one to rule them all ?
Digi-Key
Mechatronics in Design
Advances in wireless speaker performance and technology
Design Ideas
Teardown; the ultimate Consumer Product ?
Product roundup
Tales from the Cube

EDNE December 2012