Years ago in biology classes, we were taught that the brown euglena was an animal and the green euglena was a plant. The single-celled green euglena synthesized energy from chlorophyll but was otherwise pretty much identical to the brown euglena. It just goes to show how arbitrary taxonomy can be. In those days, only two kingdoms-plant and animal-were available to assign things to. Later, categorization got more complicated, perhaps to deal with quandaries like the euglena. Eventually the number of kingdoms rose to four. While obvious similarities in appearance make some classifications easy, others are more problematic. The connection between amphioxus and us is quite a reach. Categorization is externally imposed and categories will change over time.

Cores represent just such a conundrum. A core is often delivered in the form of synthesizable HDL source code resembling C, which you can implement in silicon in various ways. If the core is small enough, you can prototype your product in an FPGA. Alternatively, you may partner with an ASIC vendor. Or, if you have enough design expertise in house, you may work directly with a foundry. But is a core a chip or is it software?

This ambiguity is part of the evolutionary trend toward higher integration that's been going on since the invention of the transistor. Gone are metal chassis with vacuum tubes and actual wires. Practically gone are single-layer PCBs populated with through-hole components. Gone also are 16K DRAMs, quad NAND gates, and 8080s in 40-pin packages. Economic forces are too strong to stop this trend. Reliability, manufacturing cost, and power consumption all drive miniaturization. We're heading toward single-chip systems.

Much of the infrastructure for single-chip systems is being put into place. Semiconductor technologies continue to shrink. LSI Logic has recently announced a 0.18-micron process that will offer 26 million usable logic gates on a 20mm by 20mm piece of silicon. Since single-chip systems are semi-custom implementations, they only make sense in very high volume applications where the non-recurring engineering (NRE) costs can be distributed over a big production run.

Wide acceptance will require better development tools that can accommodate this complexity. Single-chip systems also require a lot of predesigned blocks because no one design team is going to fill up 26M worth of gates and still get to market during the same millennium. Companies like Sand, Phoenix Technologies, and others offer cores for functions like PCI, USB, Firewire, and MPEG decoding. Some companies offer processor cores. Mostly this business is in the missionary phase, and numerous problems remain to be resolved. For example, if you have cores from two different vendors, how can you be assured that they will work together? What do you do about prototype development, and after you've figured out hardware, what do you do about software development and integration?

No matter how you categorize cores, the single-chip system will play an increasingly important role in the electronics industry. How widespread this technology becomes depends on the cost of entry. Not only are the tools expensive, but NRE costs will continue to be a significant barrier.

Biological sciences are not standing still either. A check on the web indicates that the number of kingdoms has increased to five and may be on its way to six. The euglena is no longer a plant nor an animal. Problem solved-I think.

Return to Embedded.com

Send comments to: Webmaster
All material on this site Copyright © 2000
CMP Media Inc. All rights reserved.