RapidIO Connections - April 2003
Membership: Intrinsity in the Spotlight
Intrinsity, Inc. is a fabless semiconductor company enabling the creation of more powerful, real-time, and intelligent embedded systems by bringing to market the world's fastest processing solutions. Founded in May 1997 with headquarters in Austin, Texas, and offices in Boston, England, and Japan, Intrinsity has introduced 2 GHz processors that provide embedded markets with high performance software programmable solutions to replace hardwired solutions. Intrinsity has over 80 employees and is ramping to Q4 volume production of the world's first processors to support parallel RapidIO™.
Math Intensive & Adaptive Signal Processing
Resolving signals out of noise, adapting to changing environmental conditions, and applying complex processing to arrays of dynamic data are problems found in fields as diverse as DNA sequencing, military target identification, wireless communications, facial recognition, medical diagnostics, and HDTV processing.
These and many other math-intensive applications are becoming increasingly complex and adaptive-pushing today's performance ceiling. Real-time features require significantly more math performance.
The FastMATH™ Adaptive Signal Processor™ is designed to meet the needs of algorithms that must quickly process multiple data streams and adapt to environmental changes. Parallel data processing at 2 GHz is combined with the efficient algorithm partitioning enabled by a scaleable, high-performance I/O system. Unlike hardwired solutions or radical new architectures, FastMATH uses the proven MIPS® programming model and software tools to enable designers to quickly react to changing conditions and standards.
The FastMATH processor includes sixteen math processing elements that allow native computation of both vector and matrix data types. This SIMD coprocessor is tightly coupled with the MIPS scalar pipeline and architected to simplify the software work required to implement high-performance signal processing algorithms. At 2 GHz, the FastMATH processor can deliver 32 billion multiply-accumulate operations per second (64 GOPS). On a 1024-point radix-4 FFT application, the processor can sustain over 551,000 FFT's per second. In order to keep up with this high compute bandwidth, the FastMATH processor implements two 8 bit parallel RapidIO ports to provide 4 GByte/s of aggregate I/O bandwidth.
Networking Applications
As demand for more intelligent line cards increases, designers are turning to programmable solutions to improve both flexibility and functionality. Higher-layer OSI stack processing is being distributed down to line cards, driving the need for very high-performance general-purpose processors.
In addition, the increasing capacity requirements needed to process at higher network rates are pushing the need for greater performance density in a given rack volume.
The FastMIPS processor is simply the fastest MIPS32 processor and is pin-compatible with the FastMATH device. Many general-purpose embedded applications benefit from the 1MByte of on-chip memory (L2 Cache or SRAM) and the scaleable, high-performance, 4GByte/s RapidIO sub-systems.
Both the FastMIPS and FastMATH products rely on RapidIO interfaces to provide the high bandwidth data communications necessary to keep up with the computational capabilities of the processors. The selection of RapidIO for high-speed interconnect has also benefited Intrinsity customers who require scalable performance and robust features for data integrity in high-speed, complex systems. Intrinsity has demonstrated full-speed RapidIO functionality in FastMATH and FastMIPS samples and will begin full production in Q4 of this year.
