One thing is for sure. Advances in the digital realm continue at a breakneck pace, even if various limits force the move to multicore and lower-power devices.
ASICs and high-performance processors are pushing 45 nm, but not without challenges. Adoption of the latest and greatest is slowing due to technical challenges and the economy. Likewise, existing technologies are meeting the needs of most designers who are not pushing the proverbial envelope. Look for FPGAs to replace more ASIC designs even as power conservation becomes the major issue.
Improvements in electronic-system-level (ESL) design could improve the situation. Movement to tools such as SystemC is occurring, as is the use of more third-party intellectual property (IP).
MULTICORE MICROS • The multicore micro trend continues at the top end with four cores being the norm these days, while eight-core chips wait in the wings. Employing highspeed serial interconnects such as AMD’s use of Hyper- Transport and Intel’s Quick Path Interconnect (QPI) is leading to larger collections of multicore chips.
Even large numbers of cores are cropping up in chips like Tilera’s TILEPro (Fig. 1), which contains multiple 32-bit cores and a flexible distributed caching system (see “Are You Migrating To Massive Multicore Yet?” at www.electronicdesign.com, ED Online 19976). Single-chip, multicore solutions are taking on applications that had required ASICs, FPGAs, or high-performance processors. Arrays of multicore chips force programmers to learn how to approach problems from a new perspective.
The challenge for multicore will be in the software. Symmetrical multiprocessing (SMP) or non-uniform memory access (NUMA) architectures will attempt to mitigate the programming issue. However, other types of architectures like XMOS’s XS1-G4 will require new programming techniques. More radical architectures such as Intellasys’ S40C18 Seaforth (see “Multicore And Soft Peripherals Target Multimedia Applications,” ED Online 16231) will become an even greater challenge.
This is the year that GPUs turn into more general-purpose platforms. Though graphics will remain their primary purpose, GPUs aren’t just for video games any more. For example, Nvidia’s GTX 280 graphics adapter can handle more than video output using Compute Unified Device Architecture (CUDA) applications (Fig. 2).
There’s no longer a discussion of whether 8- or 16-bit processors will be squeezed out even as 32-bit platforms expand their coverage. Low power is king and the smaller platforms still have the edge. Watch while vendors with these types of multiple platforms hide their differences with C runtimes and common development tools.
FPGAS EVERYWHERE • FPGAs from Actel for less than a buck (see “FPGA Costs Half A Buck,” ED Online 20190) mark one end of the spectrum while Achronix’s Speedster 1.5-GHz FPGA marks the other (see “1.5-GHz FPGA Takes Clock Gating To The Max,” ED Online 19952). That’s a rather broad reach for one technology.
FPGA technology continues to increase gate count and hertz while cutting power requirements. This year, however, some of the most interesting aspects of the FPGA arena will occur in the software realm. Opening FPGAs to a wider group of developers is important, and vendors have been working on this area for years. These projects will bear lots of fruit.
A FLASH IN THE MEMORY PAN • Flash memory, the mainstay of microcontrollers, is bucking to replace hard drives in all but the high-capacity applications such as DVRs. In fact, this is now possible, though the economics get in the way.
Large flash storage capacity opens up microcontrollers to heftier operating systems like Linux and Windows CE, and it provides greater logging capabilities and support for larger applications. Alternate technologies like MRAM are readily available, but are still working to gain market and mind share.
While the economy may be stumbling along, don’t expect technology to take a backseat.