Air Traffic Control

The Kernel Machine

AIR TRAFFIC CONTROL: How many more air disasters?

Nigel Cook, Electronics World, January 2003, p12

In July last year, problems with the existing system were highlighted by the tragic death of 71 people, including 50 school children, due to the confusion when Swiss air traffic control noticed too late that a Russian passenger jet and a Boeing 757 were on a collision path. The processing of extensive radar and other aircraft input information for European air space is a very big challenge, requiring a reliable system to warn air traffic controllers of impending disaster. So why has Ivor Catt's computer solution for Air Traffic Control been ignored by the authorities for 13 years? Nigel Cook reports.

In Electronics World, March 1989, a contributor explained the longterm future of digital electronics. This is a system in which computers are networked adjacently, like places in the real world, but unlike the internet. An adjacent processor network is the ingenious solution proposed for the problem of Air Traffic Control: a grid network of computer processors, each automatically backed-up, and each only responsible for the air space of a fixed area. Figure 1 shows the new processing system, the Kernel computer, as proposed for safe, automated air traffic control.

Figure 1: How an adjacent network of processors would manage the ATC of European air space

This system is capable of reliably tracking a vast air space and could automatically alert human operators whenever the slant distance between any two adjacent aircraft decreased past the safety factor. Alternatively, if the air traffic controllers were busy or asleep, it could also send an automatic warning message directly to the pilot of the aircraft that needs to change course.

The existing suggestions are currently based on software solutions, which are unsatisfactory. For such a life-and-death application, there is a need for reliability through redundancy, and a single processor system does not fit the bill. System freezes must be eliminated in principle. Tracking aircraft individually by reliably using radar and other inputs requires massive processing, and a safe international system must withstand the rigours of continuous use for long periods, without any software crashes or system overheat failure.

The only practicable way to do this is through using Ivor Catt's adjacent processor network.

Originally suggested for a range of problems, including accurate prediction of global warming and long-range weather, the scheme proposed by Ivor was patented as the Kernel Machine, an array of 1,000 x 1,000 = 1,000,000 processors, each with its own memory and program, made using wafer-scale integration with 1000 silicon wafers in a 32 by 32 wafer array. The data transfer rate between adjacent processors is 100 Mb/s.

Ivor Catt's original computer development is the Catt Spiral (Wireless World, July 1981), in which Sir Clive Sinclair's offshoot computer company, Anamartic, invested £16 million. Although revolutionary, it came to market and was highly praised by electronics journals. The technology is proven by the successful introduction in 1989 of a solid-state memory called the Wafer Stack, based on a Catt patent. This received the `Product of the Year Award' from the U.S. journal Electronic Products, in January 1990.

It is a wafer scale integration technology, which self-creates a workable computer from a single silicon wafer by automatically testing each chip on the wafer, and linking up a spiral of working chips while by-passing defective ones. This system is as big an advance as the leap from transistor to compact IC (which was invented in 1959), because the whole wafer is used without having to be divided up into individual chips for separate testing and packaging.

By having the whole thing on a single silicon wafer, the time and energy in separating, testing, and packaging the chips was saved, as well as the need to mount them separately on circuit boards. By the time Catt had completed his invention for wafer scale integration, he was already working on the more advanced project, the Kernel Machine.

In the Sunday Times (12 March 1989, p. D14), journalist Jane Bird interviewed Ivor Catt and described the exciting possibilities: "in air traffic control, each processor in the array could correspond to a square mile of airspace... weather forecasters could see at the press of a button whether rain from the west would hit Lord's before the end of cricket play."

The Kernel machine versus P.C. thinking

The primary problem facing the Kernel Machine is the predominance of single-processor computer solutions and the natural inclination of programmers to force software fixes on to inappropriate hardware.

Ivor Catt has no sympathy with ideas to use his Kernel Machine for chemistry or biology research. However, this sort of technology is vital for simulation of all real-life systems, since they are all distributed in space and time. Chemical molecule simulation for medical research would become a practical alternative to brewing up compounds in the lab, if such computers became available. It would help to find better treatments for cancer.

Modern research on the brain shows that the neurons are interconnected locally. Quite often the false notion is spread that the neocortex of the brain is a type of `internet'. In reality, the billions of neurons are each only connected to about 11,000 others, locally. The network does not connect each cell to every other cell. This allows it to represent the real world by a digital analogue of reality, permitting interpretation of visual and other sensory information. Each processor of the Kernel Machine is responsible for digitally representing or simulating the events in a designated area of real space. Certainly, the Kernel machine would be ideally suited to properly interpret streamed video from a camera, permitting computers to `see' properly. This would have obvious benefits for security cameras, satellite spy and weather video, etc.

Catt filed patents for the Kernel Machine in Europe (0 366 702 B 1, granted 19 Jan 1994) and the U.S.(5 055 774, granted 8 Oct 1991), a total patenting cost around the world of about £10,000. His earlier invention, the Catt Spiral, was patented in 1972 but only came to market 17 years later after £16 million of investment by Anamartic Plc.

Patented design for the new kernel computer

Figure 2 shows how the Kernel patent differs from the Spiral in two important ways. The Spiral design as utilised in the Anamartic memory wafer, once it has been manufactured like an ordinary silicon wafer, is set up as a whole wafer computer by sending test data into a chip on the edge of the wafer.

If that chip works, it sends test data into another adjacent chip, which in turn repeats the process: sidestepping faulty chips and automatically linking up the good chips into a series network. Each chip that works is therefore incorporated into a `Spiral' of working chips, while each defective chip is bypassed. The result saves the labour of dividing up the wafer, packaging the individual chips separately, and soldering them separately on to circuit boards. It saves space, time, and money.

The problem with the Catt Spiral is that by creating a spiral or series connected memory, it causes time delays in sending and receiving data from chips near the end of the spiral. Data can also be bottlenecked in the spiral. The invention was innovative, and won awards; yet by the time Sir Clive Sinclair was ready to begin production for a massive wafer scale plug-in memory for computers, Ivor Catt was already arguing that it was superseded by his later invention, the Kernel machine. Born in 1935, Cambridge educated Catt is extremely progressive. His immediate replacement of earlier patents of his own when new developments arrive seems logical to him, although it can disturb those who invested in the previous design which has yet to make a profit.

The adjacent linking of chips into a two dimensional array in the Kernel Machine, is so-named from the `kernels' in the corners of each chip which allow networking through the chip even if it has errors and is not used itself. Kernel computers are designed to have enough networking to avoid all of the problems of the Spiral wafer. Kernel's built-in `self repair' works by ignoring individual chips when they burn out, the concept of reliability through redundancy. There are sufficient spare chips available on each wafer to take over from failures.

Catt's intended scientific and commercial computing company calls for a three-stage investment of £0.5m, £8m, and £12m, respectively. The project outline states: "The scientific market and the commercial market need to be aware that there are two fundamentally different methods of computing: large, single processing engines performing single tasks one at a time, and parallel systems where there is an array of smaller engines that perform a series of tasks independently of each other until they are brought together by some management mechanism to give a result. The scientific market's major application areas are: scientific and engineering computing; signal and image processing and artificial intelligence.

" In the commercial world there are a number of application areas where the application of very fast numerical processing is extremely useful. As the limits of physical performance are now in sight for semiconductors, the next level of performance will be achieved by applying an array of processors to a particular task. To achieve even better price/performance ratios than is presently available, the architecture needs to be flexible enough to use any one of a number of computer processor types.

"Having proven the technology and its ability to be applied to specific operational areas, the company will set to licence the technology within these application areas. The company will also develop intermediate and peripheral products on its route to the major goal; that of a parallel processing super-computer using patented technology.

"In common with all companies first entering a high technology market, this company will make a loss during the initial stages. The various stages of product development will be interposed with the marketing of that development. It is anticipated that this will reduce the negative cash flow impact inherent in an R&D environment. Industry norms have been applied to the cost of sales, marketing and administration expenditures, and to the capital costs."

In order to develop the software for the Kernel Computer, current computer technology will be used, networked in the Kernel adjacent processor array. Software, for all of the challenges facing the Kernel Computer, can be tested and debugged on this inexpensive mockup. The next phase will be the production of the first large scale super-computers using the Kernel system of wafer-scale integration.

Catt comments: "The first Wafer Scale Integration (WSI) product, a solid state disc called Wafer Stack, came to market in 1989, based on `Catt Spiral'. We can now advance to a WSI array processor, the Kernel machine, with one million processors giving one million million operations per second. The Kernel machine, when built from an array of 100 wafers, will retail for £500,000. The external control system maps out the good and bad chips, and devises a strategy for building a volatile, perfect square two-dimensional array of 1,000,000 processing elements (PE's) out of a larger, imperfect array. Reliability is achieved through redundancy; having spare PEs available.

"The project costs £20 million spread over four years. A proper figure for profit in this market would be 20% of retail price. The $0.2Bn turnover needed to justify the Kernel project is dwarfed by the $50Bn world computer market." The Kernel array computer is the machine of the future, replacing the single processor von Neumann machine of the present day.

published in ELECTRONICS WORLD January 2003   pp12, 13, 14