For more than 20 years, the technology that has dominated headlines and transformed hundreds of millions of people into computer users has been the microprocessor. This is a single chip that has circuitry to perform a wide variety of tasks. An early example was the Intel 8088 microprocessor that powered the first IBM Personal Computer in 1981. The various Pentium chips now on the market are far more powerful enhancements of this basic technology.
These chips are designed to receive a series of instructions, one at a time, and execute them accordingly. There are a great many functions available on the chip, but with few exceptions, each instruction activates only one function in turn. A Pentium performs each of these functions so quickly that it may appear to be doing several things at once. But what is really occurring is that at any given moment, only a very small portion of the chip is performing useful work. The rest of the functions built into the chip are on hot standby, consuming power but waiting to be called to duty.
This technology has made astounding progress under the leadership of Intel, AMD, Motorola, Compaq (DEC), and other firms. Microprocessors are now found in systems far larger than the PCs in which they first flourished. Twenty years ago, the most powerful supercomputers from firms such as IBM and Cray used specialized processors to calculate at very high speed. Today, they use large arrays of microprocessors to achieve their speed. While each chip still processes one instruction at a time, 1,000 chips can process 1,000 useful instructions at the same time if programmed properly.
Microprocessors are not the only way to compute. Sometimes an application does not need the wide variety of functions available on a Pentium. It may be less expensive to produce a chip with only those specific functions needed for a particular application. For example, the chip in a pocket calculator does not need circuitry to communicate at high speed over the Internet. So it is cheaper to build an application specific integrated circuit (ASIC) that meets the exact needs of the calculator. By optimizing the circuitry for only those functions needed in the target environment, it can actually be faster than a Pentium when performing those functions.
Within a personal computer, for example, you may find specialized chips—ASICs—that support the CD-ROM drive, the video display, the sound system, or networking with other PCs.
Both microprocessors and ASICs share a common trait. Their functions are built into the chip at the time of manufacture. They are hard-wired into the chip.
For more than a decade, firms such as Altera and Xilinx have been developing an alternative type of chip called a field programmable gate array (FPGA). These chips leave the factory with no functions wired. From the same production run, one batch of chips may go to an appliance manufacturer where they are configured to control dishwashers, and the next batch may go to a television manufacturer where they are configured to control the picture contrast.
What these manufacturers are really doing is using FPGAs as ASICs. They design a circuit with specific functionality for a particular application, i.e., an ASIC. But the one-time costs of ASIC manufacture can be prohibitively expensive. So instead they buy a batch of FPGAs and configure them with their ASIC design. For a small production run, this turns out to be less expensive than producing a similar number of ASICs. If the production run is large, that same manufacturer may choose to go directly to ASIC fabrication instead.
Early FPGAs were designed to be programmed only once after they left the factory. And that is essentially how most FPGAs are still used today. In our appliance example, the control circuitry may be saved in a read-only memory (ROM) chip that is soldered onto the dishwasher control system along with the FPGA. The circuitry is copied from the ROM to the FPGA, which never receives any other configuration in the user environment.
But today’s most advanced FPGAs are not limited to a single configuration operation. They are designed to be reconfigured as needed by the user. It is possible to configure the chip to perform one function, use it for that purpose, then erase that configuration and configure the circuit in a different way for another function. With very few exceptions, however, this property is not being used today.
Star Bridge Systems has developed a system to exploit the advantages of FPGAs to overcome the limitations of traditional microprocessors. In the process, it has developed related technologies that are applicable in a wide range of business applications.
The core technology developed by Star Bridge is that of reconfigurable computing. This term, coined by the company’s co-founder, Kent L. Gilson, refers to the capacity of a state-of-the-art FPGA to be configured, perform the function desired, and then be configured again to perform a different function. As the functions needed from the FPGA change, the configuration of the FPGA is changed to meet them.
Recall that on a microprocessor, only a small portion of the chip is in use at any one time, while the rest of the chip is in hot standby mode. The hundreds of functions embedded in the circuitry are sitting idle, as they cannot be called until the one function currently in use has completed its work.
Using Star Bridge technology, however, the circuitry needed at the moment is configured on a small portion of the FPGA chip. But instead of leaving the rest of the FPGA in hot standby mode, we use it to perform many other functions at the same time.
Also recall that a supercomputer is really an array of 1,000 or more microprocessors, each doing one function at a time, with the remainder of each chip in hot standby mode. Using the Company’s technology, you could instead run perhaps 100 or more processes on a single FPGA. In a similar way, you could replace the 1,000 microprocessors in a supercomputer with a far smaller number of FPGAs.
Star Bridge Systems has developed the hardware necessary to coordinate the functioning of multiple FPGAs in a single computer. Star Bridge has also developed the software needed to configure each FPGA to perform multiple functions at the same time.
The reconfigurable computing system described above has obvious advantages (which will be detailed below). Surely there must be other people who can develop this type of technology.
Yes, there are others who have thought about this concept. The advantages of reconfigurable computing have been theorized for some time. Federico Faggin, principal developer of the microprocessor, has said:
“[A]s time goes on, reconfigurable chips will have an ever increasing impact in mainstream applications. . . . Then the very raison d’etre of the traditional computer disappears. . . . I believe the natural evolution of this remarkable new technology will augment and deeply transform the architecture and capabilities of future computers.”
But it is one thing to conceive of space travel and an entirely different matter to put Neil Armstrong on the moon. The technical difficulties to be overcome by any developer of an effective reconfigurable computer are enormous. Indeed, Star Bridge Systems has not solved all of the problems involved. But the Company has solved a great many of them, and it has a clear idea of the problems that remain and where solutions might be found. We know of no other company that is developing reconfigurable computing in a fashion similar to ours.
It should also be understood that beyond having an idea and the resources to develop it, a company must be motivated to go down this path. Building supercomputers today is a very profitable business, with only a handful of companies having the resources to develop them. The modern supercomputer consists of scores of refrigerator-sized cabinets crammed with microprocessors and related circuitry. The systems cost tens of millions of dollars, and may have very profitable multi-year maintenance contracts as well.
Let’s imagine that an engineer working for a supercomputer manufacturer came up with a suggestion to start a multi-million dollar research and development effort to replace each of the company’s $20 million supercomputers with a single cabinet the company could sell for $2 million. Would the senior management of the company welcome that suggestion?
In fact, it should. Because otherwise a little company called Star Bridge Systems might derail the gravy train and disrupt the business with its revolutionary technology.
The advantages of the Star Bridge technology are only hinted at in the preceding discussion. Each deserves further explanation.
Star Bridge reconfigurable computing technology is not limited to supercomputers. It is equally applicable to the full range of digital processing systems in use today or envisioned for tomorrow: wristwatch video recorders, smart cell phones with Internet access, artificial vision systems embedded in a blind person’s body, satellite telecommunications servers, automotive collision avoidance and navigation systems, desktop PCs, Internet servers, or medical imaging and interpretation systems. If a system uses digital processing, the odds are excellent that it could be enhanced with this technology.
For decades, technological advances have roughly doubled the power available in a microprocessor every 18 months, in a process commonly called Moore’s Law. This has brought enormous increases in the power purchased per dollar, and the trend continues. Amazingly, FPGA companies such as Xilinx are increasing the density of system gates (circuitry) on an FPGA at an even faster pace:
1998 1 million gates per FPGA
1999 2 million gates per FPGA
2000 4 million gates per FPGA
2002 10 million gates per FPGA
2004 50 million gates per FPGA
This means that more power will be available in fewer FPGA chips on a continuing basis over the next several years, and history suggests that the price per gate will continue to decline at a rapid pace.
If you could design a system architecture based upon microprocessors (Moore’s Law growth curve) or FPGAs (Xilinx growth curve), which would seem more cost-effective over the long term?
In a state-of-the-art supercomputer using thousands of microprocessors, a very small portion of each chip does productive work at any moment. Yet the entire chip consumes power. The Star Bridge reconfigurable computing approach uses fewer chips, but a much larger proportion of the circuitry of each chip does meaningful work at any moment.
The power consumption problem is becoming a significant one. It is easy to think of a microprocessor as consuming just a small amount of power. And in the great scheme of things, that is still true. But put 1,000 microprocessors in the same system, and the power consumption becomes a factor, for the chips, their supporting circuitry, the air conditioning system, etc. Much has been written recently about the lack of power generating capacity in California, and the inability to meet current demand. With power companies now scheduling brownouts and blackouts, there is serious concern whether any local power company can commit to providing the amounts of power that a supercomputer center needs for a new system based upon conventional microprocessor technology.
The IBM Pacific Blue supercomputer uses 5,856 microprocessors and consumes 3,900,000 watts of power. The Star Bridge HAL-300 uses 180 FPGAs and consumes 1600 watts of power. While we are not claiming the two machines have equivalent power today, they are a lot closer in capability than the 2,000 to 1 ratio of power consumption would indicate.
When a conventional computer system is put into operation, its development is frozen in time. New functionality can only be added by replacing the ASIC or microprocessor with newer chips having the desired features.
With Star Bridge technology, however, it is possible to change the functionality of the system through software. The benefits of this approach can perhaps best be demonstrated by considering the problem of technological change in the telecommunications industry.
Let’s assume that a satellite television signal provider puts the latest state-of-the-art technology in a new broadcast satellite 22,000 miles above the earth. Then the technology changes, and the functionality of the computers aboard the satellite must be changed. With conventional systems, there is no economical way to swap out the circuitry on a device in high orbit, far beyond the range of the Space Shuttle. But with Star Bridge technology, the appropriate software can be transmitted to the satellite to reconfigure the hardware as needed.
The same idea could apply to ground-based systems as well. With reconfigurable circuitry built into a television set, for example, new features could be programmed simply by embedding the appropriate software in the broadcast of the evening news. In moments, 10 million TVs could be upgraded with new features without a single house call by the repairman.
The not-so-secret-weapon that allows Star Bridge to plan these advanced technologies is called Viva. This is the company’s first completed software product and the most advanced tool anywhere for programming applications in silicon chips.
Viva 1.0 empowers application developers chip designers
to develop new data processing applications and enhance existing applications
with greater ease, speed, and economy.
This means that with Viva 1.0 and its progeny, it will be possible to
improve the performance of computers of all sizes and the performance of chips
embedded in products of all kinds—consumer electronics, communications devices,
household appliances, automobiles—more quickly and at less cost than with
conventional software development tools.
Viva 1.0 demonstrates that Star Bridge technology offers potential benefits in every design parameter governing the development of data processing systems—performance, size, cost, power consumption, reliability, and time to market.
A
superior development tool like Viva 1.0 can become profitable in many
markets across the entire spectrum of information technology and electronics. Almost anything with a chip inside—from scientific supercomputers and high-end servers to
personal
digital assistants—can
be improved by using this technology.
With Star Bridge technology, full use of every potential advantage of FPGAs becomes possible for the first time. The inherently parallel nature of these chips, coupled with the power of Viva software to reconfigure them, will allow multiple functions to operate in a single chip that might otherwise require numerous special purpose chips. Viva reconfigures the FPGAs in a Star Bridge Hypercomputer to perform every required function, and multiple tasks may be executed simultaneously.
This ability to redesign the circuitry of chips and silicon-based devices, together with the enormous speeds achievable within a Hypercomputer, give only a hint of the powerful technology being developed by Star Bridge Systems.
Star Bridge Systems
7651 South Main Street
Midvale, UT 84047
Phone 801-984-4444
Fax 801-984-4445
www.starbridgesystems.com