Embedded Computation

Transcription

1 Embedded Computation

2 What is an Embedded Processor? Any device that includes a programmable computer, but is not itself a general-purpose computer [W. Wolf, 2000]. Commonly found in cell phones, automobiles, airplanes, basestations, satellite systems, and home appliances. Typically programmed once to perform specific tasks, but need to be flexible enough to handle similar tasks. Often take advantage of application-specific characteristics to optimize the processor design. Require low cost and power, with high reliability and performance.

3 Embedded Processors Embedded processors are everywhere!

4 Attributes of Embedded Devices Not designed to be general purpose Laptops, desktops, and workstations are not embedded Processor not visible to the user and user rarely or never loads a program Rarely require backward compatibility or upgradability Usually designed for a small set of tasks Can be highly optimized for these tasks Often have hard constraints on cost, power, energy efficiency, reliability, and size Often have real-time constraints What are examples of embedded devices?

5 Types of Compatibility Binary compatibility Run previously compiled binary programs Often required in general purpose p devices Less of an issue in embedded devices What might be some exceptions? Embedded compatibility constraints Feature compatibility Source compatibility Tool compatibility OS compatibility

6 Insects of the Computer World Like insects, embedded processors come in all shapes and sizes

7 Types of Embedded Processors Computational micros (32- or 64-bit datapaths) CPU of workstations, PCs, or high-end portable devices (PDAs) x86, PA-RISC, PowerPC, SPARC, etc. Embedded general purpose p micros (32-bit datapaths) Designed for a wide range of embedded applications Often scaled-down version of computational micros ARM, PowerPC, MIPS, x86, 68K, etc. Microcontrollers (4-, 8-, or 16-bit datapaths) Integrate processing unit, memory, I/O buses, and peripherals Often low-cost, high-volume devices Domain-specific processors (datapath size varies greatly) Designed for a particular application domain Digital signal processors, multimedia processors, graphics processors, network processors, security processors, etc.

8 8-bit Processors Dominate

9 Processor Market 2001 processor market by volume: Computational micros: 2% Embedded general-purpose micros: 11% DSPs: 10% Microcontrollers: 80% 2001 processor market by revenue: Computational micros: 51% Embedded general-purpose micros: 8% DSPs: 13% Microcontrollers: 28% Hi h th t d f b dd d i Higher growth expected for embedded micros, DSPs, and microcontrollers

10 Growing Demand Embedded processors account for Over 97% of total processors sold Over 60% of total sales from processors Sales expected to increase by roughly 15% each year

11 Why are there so many? Average middle-class American home has 40 to 50 embedded processors in it Microwave, washer, dryer, dishwasher, TV, VCR, stereo, hair dryer, coffee maker, remote control, humidifier, heater, toys, etc. Luxury cars have over 60 embedded processors Brakes, steering, windows, locks, ignition, dashboard displays, transmission, mirrors, etc. Personal computers have over 10 embedded processors Graphics accelerator, mouse, keyboard, hard-drive, CD-ROM, bus interface, network card, etc.

12 Characteristics of Embedded Processors Sophisticated functionality Complex algorithms - operations may be very sophisticated (e.g., data compression) Complex interface - often control complex user interfaces (e.g., GPS display) and can be I/O intensive Safety and reliability - often have hard reliability and correctness constraints (e.g., ABS system) Meeting deadlines real time - have to perform in real time; if data isn t ready by a certain deadline, the system breaks multi-rate - several real time activities going on at the same time (e.g., multimedia systems)

13 Characteristics of Embedded Processors Power, costs, and other limitations power - power consumption directly affects hardware costs and battery life (portable systems) manufacturing costs - often used in low cost applications (e.g., microwaves) memory space - limited amount of on-chip memory and no external disk product life cycles - life cycles often measured in months (e.g., cellular l phones) time to market - produced by small teams working on a tight deadline g What are other characteristics?

14 Domain-specific Embedded Processors Much embedded processor research focuses on domain-specific processors, such as Digital signal processors Multimedia processors Graphics processors Network processors Why is this a research focus? The processors have special requirements for their particular application domain Leads to specialized feature for their application domain

15 Domain-Specific Embedded Processors erformanc ce Power/P FFT Processors MPEG Processors FIR Processors Embedded Application-Specific Processors Graphics Processors DSP Processors Network Processors Embedded Domain-Specific Processors Workstations Personal Computers General-Purpose Processors Programmability and Flexibility

16 Embedded DSP Processors Embedded DSP processors are optimized to perform DSP algorithms; speech coding, filtering, convolution, fast Fourier transforms, discrete cosine transforms y N = k bn x k n n= 0 DSP processors feature Deterministic i i execution times Fast multiply-accumulate instructions Multiple data accesses per cycle Specialized addressing modes Efficient support for loops Operate on streaming data

17 Texas Instruments TMS320C62xx N. Seshan, High VelociTI processing [Texas Instruments VLIW DSP architecture], IEEE Signal Processing Magazine, v. 15, no. 2, pp , 117, 1998.

18 TigerSharc DSP J. Fridman and Z Greenfield, The TigerSHARC DSP architecture, IEEE Micro, vol. 20, no. 2, pp , 2000.

19 DSP vs. General Purpose Processors DSP Processors Deterministic Execution Time Loop Buffers Multiple l Operations per Instr. Operate on Stream Data Integer & Fractional Arithmetic Very Low Power Fast Interrupt Processing Static Scheduling in Software Programmed in assembly Scratchpad memories General Purpose Processors Non-Deterministic Execution Time Branch Prediction Multiple l Instr. per Cycle Caches Assume Locality Integer and Floating-Point Arithmetic Much Higher Power Slow Interrupt Processing Dynamic Scheduling in Hardware Programmed in high-level language Instruction and data caches General-purpose processors have added more DSP-like features and DSPs have added some general-purpose capabilities

20 Emerging DSP Architectures Parallelism at multiple levels Multiple processors System-on-a-chip designs Multiple simultaneous tasks Multithreaded processors Multiple instruction per cycle Very Long Instruction Word (VLIW) architectures Very Long Instruction Word (VLIW) architectures Multiple operation per instruction Single Instruction Multiple Data (SIMD) instructions Architecture/compiler pairs improve performance and help manage application complexity

21 Very Long Instruction Word Architectures Each long instruction performs multiple operations in parallel Branch Memory Memory Arithmetic Logic Vector Needs a good compiler that understands the architecture Allows deterministic execution times Code growth can be reduced by allowing operations within an instruction to be performed sequentially a given field to specify different types of operations Seq Branch/Mem Mem/Arith Arith/Logic Vector

22 Embedded Multimedia Processors Embedded multimedia processors operate on different types of data simultaneously Video, images, audio, and text Typical multimedia applications include Video and image compression Video and image processing Speech and audio coding Multimedia characteristic Little data reuse streaming data Low precision data 8 or 16 bits Lots of data parallelism Very computational intensive

23 Imagine Multimedia Processor B. Khailany et al., Imagine: media processing with streams, IEEE Micro, vol. 21, no. 2, pp , 2001.

24 SIMD Instructions Recent multimedia processors commonly support Single Instruction Multiple data (SIMD) instructions The same operation is performed on multiple data operands using a single instruction A3 A2 A1 A0 B3 B2 B1 B0 A3+B3 A2+B2 A1+B1 A0+B0 Exploits low precision and high data parallelism of multimedia applications

25 SoC Design System-on-Chip (SoC) designs integrate microprocessors, DSPs, memories, buses, and peripherals on a single chip Years Process Technology 0.35 micron 0.25 micron 0.18 micron 0.13 micron Cost of Fab 1.5 to 2.0 B$ 2.0 to 3.0 B$ 3.0 to 4.0 B$ 4.0 to 5.0 B$ Design Cycle 12 to to 12 8to10 6to8 months months months months Design K 1-2M gates 4-6M gates 8-10M gates Complexity gates L Todd and A. McNelly., The Transition to System-on-a-Chip, Technical Report, Cadence Design Systems, February, 2001.

26 Future Embedded Systems Future embedded systems will integrate Multiple heterogeneous embedded processors Instruction & data memories Instr Data Network interfaces Memory Hardware accelerators All on the same chip 16 Voice/HiFi DSP Important research issues System integration Hardware/software codesign (Re)configurable systems Performance & power analysis Testing and fault tolerance System security CODECS Intf Micro Controllers Memory FFT USB Port RS HS-DMA DSP DSP AFE Intf PLB/OPB Bridge Code 2k x 24 Data 8k x 16 DSL Utopia-2 PCI Bus Master Data 10/100 Mbps FENET MAC 10/100 Mbps FENET PHY

27 Conclusions Embedded processors are ubiquitous Demand for systems that use them is increasing Domain-specific embedded processors are specifically tailored to a particular application domain Needed to meet stringent power, performance, cost, and real-time processing constraints Challenging research problems in embedded processor design Reducing power consumption Selecting appropriate hardware resources Performing hardware/software co-design Ensuring reliability and security