With numerous kinds of processors with various design philosophies available at our disposal for using in our design, following considerations need to be factored during processor selection for an embedded system.
- Performance Considerations
- Power considerations
- Peripheral Set
- Operating Voltage
- Specialized Processing Units
Now let us discuss each of them in detail.
Performance considerations:
The first and foremost consideration in selecting the processor is its performance. The performance speed of a processor is dependent primarily on its architecture and its silicon design. Evolution of fabrication techniques helped packing more transistors in same area there by reducing the propagation delay. Also presence of cache reduces instruction/data fetch timing. Pipelining and super-scalar architectures further improves the performance of the processor. Branch prediction, speculative execution etc are some other techniques used for improving the execution rate. Multi-cores are the new direction in improving the performance.
Rather than simply stating the clock frequency of the processor which has limited significance to its processing power, it makes more sense to describe the capability in a standard notation. MIPS (Million Instructions Per Second) or MIPS/MHz was an earlier notation followed by Dhrystones and latest EEMBC’s CoreMark. CoreMark is one of the best ways to compare the performance of various processors.
Processor architectures with support for extra instruction can help improving performance for specific applications. For example, SIMD (Single Instruction/Multiple Data) set and Jazelle – Java acceleration can help in improving multimedia and JVM execution speeds.
So size of cache, processor architecture, instruction set etc has to be taken in to account when comparing the performance.
Power Considerations:
Increasing the logic density and clock speed has adverse impact on power requirement of the processor. A higher clock implies faster charge and discharge cycles leading to more power consumption. More logic leads to higher power density there by making the heat dissipation difficult. Further with more emphasis on greener technologies and many systems becoming battery operated, it is important the design is for optimal power usage.
Techniques like frequency scaling – reducing the clock frequency of the processor depending on the load, voltage scaling – varying the voltage based on load can help in achieving lower power usage. Further asymmetric multiprocessors, under near idle conditions, can effectively power off the more powerful core and load the less powerful core for performing the tasks. SoC comes with advanced power gating techniques that can shut down clocks and power to unused modules.
Peripheral Set:
Every system design needs, apart from the processor, many other peripherals for input and output operations. Since in an embedded system, almost all the processors used are SoCs, it is better if the necessary peripherals are available in the chip itself. This offers various benefits compared to peripherals in external IC’s such as optimal power architecture, effective data communication using DMA, lower BoM etc. So it is important to have peripheral set in consideration when selecting the processor.
Operating Voltages:
Each and every processor will have its own operating voltage condition. The operating voltage maximum and minimum ratings will be provided in the respective data sheet or user manual.
While higher end processors typically operate with 2 to 5 voltages including 1.8V for Cores/Analogue domains, 3.3V for IO lines, needs specialized PMIC devices, it is a deciding factor in low end micro-controllers based on the input voltage. For example it is cheaper to work with a 5V micro-controller when the input supply is 5V and a 3.3 micro-controllers when operated with Li-on batteries.
Specialized Processing:
Apart from the core, presence of various co-processors and specialized processing units can help achieving necessary processing performance. Co-processors execute the instructions fetched by the primary processor thereby reducing the load on the primary. Some of the popular co-processors include
Floating Point Co-processor:
RISC cores supports primarily integer only instruction set. Hence presence of a FP co-processor can be very helpful in application involving complex mathematical operations including multimedia, imaging, codecs, signal processing etc.
Graphic Processing Unit:
GPU(Graphic Processing Unit) also called as Visual processing unit is responsible for drawing images on the frame buffer memory to be displayed. Since human visual perception needed at-least 16 Frames per second for a smooth viewing, drawing for HD displays involves a lot of data bandwidth. Also with increasing graphic requirements such as textures, lighting shaders etc, GPU’s have become a mandatory requirements for mobile phones, gaming consoles etc.
Various GPU’s like ARM’s MALI, PowerVX, OpenGL etc are increasing available in higher end processors. Choosing the right co-processor can enable smooth design of the embedded application.
Digital Signal Processors:
DSP is a processor designed specifically for signal processing applications. Its architecture supports processing of multiple data in parallel. It can manipulate real time signal and convert to other domains for processing. DSP’s are either available as the part of the SoC or separate in an external package. DSP’s are very helpful in multimedia applications. It is possible to use a DSP along with a processor or use the DSP as the main processor itself.
Price:
Various considerations discussed above can be taken in to account when a processor is being selected for an embedded design. It is better to have some extra buffer in processing capacities to enable enhancements in functionality without going for a major change in the design. While engineers (especially software/firmware engineers) will want to have all the functionalities, price will be the determining factor when designing the system and choosing the right processor.
]]>
0 Comments