The Promise of rRAM

 
Thanks for taking the time to consider rRAM. We think rRAM is a powerful new concept in silicon devices combining the strength and flexibility of standard RISC processors with a high level of integration and interconnection.

rRAM is a new kind of scalable computing component supporting a wide range of applications and ultimately offering performance orders of magnitude higher than conventional single processor systems.
 

Killer Applications

Today's PC's have a small number of microprocessors. The focus of development has been to make the main processor as fast as possible. However many applications are now appearing which need the power of distributed processing. VLIW, and DSP approaches have improved available computing power but still have significant bottle necks in their architecture. The rRAM will excel in a number of partitionable applications both for the main stream and specialized processing. Applications which will leave conventional processors standing are:

• High resolution virtual reality rendering in real time in a low cost PC. No silicon (and certainly no programmable silicon) can do that today.
• Soft, real time, MPEG or other video encoding and decoding.
• Fast real time, soft DSP applications.

The Software Question?

What about the software? This is always the first question when array processors of any kind arise. We think rRAM side steps the issue. This is our thinking:

• Porting of applications? Our work with a number of graphic (and other structured) applications shows that C code can be ported to rRAM in about 2% of the original code development time. It is mostly a process of separating inner loops and associated data and assigning a processor. The control flow and the detailed code remains largely unchanged.
• A library of commonly needed high level rRAM functions will be available to the developer. These will be host based modules which implicitly handle most of the setup and communication issues to (and within) the rRAM. The developers will link their programs to these high level functions like any other library. High level functions examples are DCT, motion estimation, template matching, wavelet etc.
• The rRAM high level library will have a host based alternative with the same interfaces so that the program can be largely debugged in a non rRAM environment.
• Porting of OS? rRAM is a very general purpose engine and we suppose it would be possible to port Linux or NT. However, that is not the forte of rRAM. Think more (for example) of replacing the graphics engine of your PC with something much more generic that can be programmed to perform a wide variety of applications.
• 
  In other words we think the programmability of the rRAM improves rather than diminishes the flexibility and programmability of the systems it is added to.

The rRAM device family can cover a wide variety of applications in the multimedia and general data processing markets, taking market share from a wide variety of toady's hardwired devices.
This can improve your market share and reduce number of different products your company builds.

Key user features of the rRAM are:

• Performance: The rRAM64 has 64 times the performance of a single processor in target applications. Making it possible to use the device where hardwired logic or FPGAs are required today.
• Flexibility: As a configurable device a wide variety of applications can be addressed with a single part.
• Easy to use: Conventional, existing, language compiler tools are used for programming. The programming model is very simple making it possible for any C programmer to write code. The complex VHDL, synthesis, layout, and timing issues of conventional FPGA or customer device development are avoided.
• Fast time to market: Because of the ease of use features rRAM code can be developed and prototyped quickly.
• Dynamic configuration: rRAM can change its personality within a few clock cycles into a completely different function.
• Field upgradable: Unlike hardwired approaches rRAM can be field upgradable allowing new multimedia standards to be accommodated or whole new applications marketed.

Although rRAM is indeed challenging to today's silicon technologies we are are focused no the long term opportunities for this architecture. By 2005 the progress of silicon technologies will be such that large numbers of processors, each with several megabytes of local DRAM can be integrated on a single chip. At this point it will be possible to place a real OS on each processor and use conventional constructs for inter process communication.

Updated: 11/11/06