1. Presented By: Rodney Fluharty Dec. 07, 2000

2. Who is ARM? Advanced Risc Microprocessor is the industry's leading provider of 16/32-bit embedded RISC microprocessor solutions. Licenses their high-performance, power- efficient RISC processors, peripherals, and system-chip designs to leading international electronics companies.

3. Who uses ARM technology? Atmel, Cirrus Logic (www.cirrus.com), Fujitsu, Mitel (www.mitel.com), IBM, LG Semicon (www.lgsemicon.co.kr/), LSI Logic, Lucent Technologies (www.lucent.com), National Semiconductor, NEC, Oki, Samsung, Seiko Epson (www.epson.co.jp/), Sharp, Texas Instruments, Toshiba, and VLSI.www.cirrus.comwww.mitel.comwww.lgsemicon.co.kr/www.lucent.comwww.epson.co.jp/

4. Why Don’t I see “ARM” chips? They are sold as VC (virtual components) and IP (intellectual property). Their designs are embedded and only the technology is ARM’s. Much cheaper to use existing technology.

5. Brief Overview of ARM 940T Member of ARM 9 family. Complete CPU subsystem (Bus, Cache, Core) Harvard Architecture: Separate Data and Instruction Memory. 31 general-purpose registers with 16 simultaneously visible. Core is a 32-bit RISC processor. Do not support Virtual Addressing.

6. Core Block Diagram

7. Microprocessor Block Diagram

8. Number Crunching Single-cycle 16X32-bit multiply- accumulate (MAC) unit Integer based only. Floating point would require a co-processor. Lack Integer divides (must synthesize division)

9. Bus Architecture/Clocking Methodologies Separate data/instruction busses. Two clock inputs (BCLK, FCLK). 3 Modes of clocking: FastBus: Used with high speed memory; BCLK controls ARM91TDMI, cache ops, AMBA Bus. FCLK ignored.

10. Synchronous - Used for low-speed memory; both clock inputs used. BCLK controls bus FCLK controls core, cache. –Rules for Synchronous »FCLK > BCLK »BCLK transition must occur when FCLK high.

11. Asynchronous - Used for low-speed memory; both clock inputs used. BCLK controls bus FCLK controls core, cache. –Rules for Synchronous »FCLK > BCLK

12. Cache Description Instruction/Data Cache = 4Kb 8 word write buffer Each cache comprises four, fully- associative 1kb segments. Single-cycle reads, one/two cycle writes (depending on sequence of instructions).

13. Cache Description Continued Implements “Read-on-miss replacement” policy. Selection by randomly clocked rows (unless locked). Can use “Write-back” or “Write-through” Implement both “Valid” and “Dirty” bits.

14. Cache Architecture

15. Pipelining 5 Stage pipeline (fetch, decode, execute, memory, write-back). Implements bubble insertion.

16. Introduction to Thumb Why waste memory on instructions if not necessary? 16 bit subset of the 32 bit instruction set. Thumb module located in pipeline. Decompresses 16 bit instruction to 32 bit equivalent with no delays. Up to 30% code density improvement.

17. A Closer Look

18. How much does this affect? 36 of the ARM’s native instructions have been adapted to Thumb technology. These did not benefit from the full 32 bit instruction.

19. Jazelle Technology Java was developed for embedded systems, so it makes sense to optomize an embedded processor for Java! Historically: – Java source code is converted to a Java byte code. –Machine had to convert byte code to instructions at execution time. –This can be very slow on low-power embedded hardware (cell phones, set-top boxes, handhelds).

20. Wake me up when it’s loaded... Original hardware solutions involved costly external co-processors. ARM’s solution: Add one more instruction and about 12,000 gates to the decode. Enter ‘BXJ Rm’ and the ARM goes into Java mode eliminating the slower JVM Certain registers are re-assigned to Java. Still ARM and Thumb compatible.

21. Let the numbers speak

22. Conclusion ARM continues to produce high quality, embedded processors. ARM has developed new technologies to optomize hardware. Newer products such as the ARM10 or StrongARM (@600MHz) will likely appear in everyday life.