EEMBC’s Automotive/Industrial Microprocessor Benchmarks June 4, 2004 EEMBC’s Automotive/Industrial Benchmark Suite

16 different used in automotive applications

• Angle-to-time • IDCT conversion • Matrix arithmetic • Basic floating point • Pointer chasing • manipulation • Pulse width • Cache buster modulation

• CAN remote • Road speed request calculation

• FFT • Table lookup and interpolation • FIR Automark score aggregates • Tooth-to-spark the individual • IIR performance measures EEMBC Adds Suite of Benchmarks for 8- and 16-bit

In the automotive world, 8- and 16-bit microcontrollers remain an important category of processing devices

• Industry’s first certifiable benchmarks for 8- and 16-bit microcontrollers

• Converted from automotive/industrial benchmarks

• Shrunk for smaller memory footprints

• No floating-point

• New task-based benchmark Why high performance microcontrollers are crucial for automotive.

• Microcontrollers help to reduce emissions and fuel consumption because they provide tighter feedback and allow the system to more rapidly and accurately track the operating environment and conditions • Though processors are already used for widespread applications, there are still a number of areas where processors can bring new capabilities and lower prices • The burgeoning performance of 16- and 32-bit processors permits using chips for new control functions, particularly for safety and entertainment applications inside the car • Additionally, for the telematics industry to grow as predicted, inexpensive yet efficient processors must perform many tasks, such as voice recognition Powertrain – Driving Forces and System Trends

2000 2005 2010 Driving experience — better engine refinement Reduction of emission levels/fuel consumption

Variable Valve Camless Engine Actuation EVTEVT Alternative PiezoPiezo power sources Direct Injection Hybrids FuelFuel CellCell

Electronic Controlled Integrated Transmission Starter/Alternat or CVTCVT AutocodeAutocode Torque based Integrated Adaptive control Powertrain Powertrain Control Control

Source: Infineon Technologies Powertrain Driving Forces for Higher Performance

50 Performance 12 Average Fuel Consumption

CO-Emission g/km 10 40 Liter / 100km / Liter

8 30 32-Bit32-Bit 6 20 Integrated 16-Bit16-Bit Direct Injection Starter/ 4 Alternator 10 8-Bit8-Bit EURO1 Emission 2 EURO4 Electronic Controlled Transmission

Performance relative to 1980 (=1) 0 0 1980 1990 2000 2005

Source: VDA, Infineon Technologies EUROx = European Emission Standards Powertrain Driving Forces for Higher Performance

Log 100000 Powertrain 32bit Audo-NG 10000 150 MHz TriCore P-BGA-416 FPU, 2MB eFlash (ECC), PCP GPTA-4, MultiCAN, TTCAN 16bit 0.13µ-40/125°C eFlash Audo TC1796 1000 40 MHz TriCore P-BGA-329 PCP, GPTA, TwinCAN -40/125°C TC1775 100 C167/XC167 Production TC 1765 P-MQFP-144, bare die 32K / 128K ROM 40 MHz TriCore Development 25 / 33 MHz XC167CI -32FF P-LBGA-260 DMA, GPTA, TwinCAN -40/125°C C167CR-xR(33)M ES 10 Q1/2004 TQFP-144 Concept C166S V2 Core 256kB eFlash (ECC) incl. 8k E²PROM ES = Final 20/40MHz Engineering Samples 1

Q = Qual finished 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 SOP = Start of Production * = not for new designs

Source: Infineon Technologies Changing Requirements for Microcontrollers

Feature 1994 2004 Architecture 16-bit 32-bit

Frequency 33 MHz 150 MHz

Instruction cycle time 60 ns 4.44 ns (typical)

Flash Memory 0 2 MB

RAM Memory 4 kB 192 kB

Input/ Output Handling Peripheral Event Controller 32-bit I/O Processor

Timer for Motor Control Capture/Compare GPTA (9 Timers) (96 Timers)

This table summarizes the information provided in the previous graph, showing the dramatic change in requirements over the past 10 years.

Source: Infineon Technologies EEMBC Benchmark Results

Silicon Benchmark Scores Mitsubishi M16C/80 - 20 MHz

NEC V850E - 50 MHz

Infineon TriCore/TC11IB - 96 MHz

0 5 10 15 20 25 30 35 Automarks This chart compares three processors using EEMBC’s Automark (The Automark is calculated by taking the geometric mean of each of the individual scores within the Automotive/Industrial benchmark suite). These results are more or less ‘expected’ because a) the Mitsubishi device is a 16-bit microcontroller compared to the other two devices which are 32-bit microcontrollers; b) operating frequency is a big factor in performance measurements; c) after normalizing for operating frequency, the Infineon device is still approximately 10% faster than the NEC device, this is most likely related to the former’s enhanced ability to perform signal processing. EEMBC Benchmark Results

Simluation Benchmark Scores

Infineon TriCore/TC1M

IBM ppc405d4v6

ARM1026EJ-S

SuperH SH4-202

SuperH SH5-103

0 0.10.20.30.40.50.60.7 Automarks per MHz

Similar to the explanation on the previous slide, this chart compares 5 processor architectures using EEMBC’s Automarks. One of the significant differences on this chart is that these scores are all based on running the benchmarks using simulators for these processors and all scores are normalized to 1MHz. Without going into the specific details for each processor, it’s important to note that the ARM and SuperH devices support hardware floating-point units that will help boost their performance on several of the Automotive/Industrial benchmarks.