Natural Motion — Melzonic

The processing power of a hundred computers on one chip

"When Gerard de Haan had completed his algorithm, his calculation method for motion estimation had to be transformed into silicon. I don't mean the fab process, but the design of the chip, the integrated circuit comprising switches, memory and signal paths. It was not an easy task because in this case it concerned a 'megachip' with a million transistors. To design such a chip successfully and without making any mistakes, you need special tools. Our group within Philips Research had just developed such a tool."

Speaking is Dr Jef van Meerbergen, an engineer who was involved in the development of Melzonic: the chip in the newest televisions that provide natural motion. "Our group's business is Digital VLSI Design," says Van Meerbergen. "VLSI stands for Very Large Scale Integration'. This means integrating or joining millions of microscopically small components on a substrate or carrier".

"Because there are so many components to be integrated, the chips are very complex. You must therefore make the development process as automated as possible. Basic blocks comprising standard cells are used for this purpose. The blocks are joined by means of a high-level computer language, and you arrange them on the surface of a chip."

The design of a chip, says Van Meerbergen, is best compared with building a town. The transistors are the houses. Clusters of these form suburbs with different functions. Between these run streets and pathways for carrying energy and information. There are traffic lights at junctions, to control the traffic.

We previously drew such 'town plans' manually on a drawing-board," continues Van Meebergen. "But they were only small towns in those days. Only a few years ago, we designed such a 'town' of 70,000 houses in 15 months. With Melzonic, a million houses were involved, and traffic speed was ten times faster. If we had designed such a thing the old-fashioned way, it would have taken 150 times as many people, or 150 times longer, or a combination of both. That is absurd of course, because there is not so much time or manpower available. That's why we work continually toward methods of design that can save time and minimize manpower requirements."  

As luck would have it, when Gerard de Haan invented his method of motion estimation, the group of Jef van Meerbergen had just developed 'Phideo': a very advanced method of design which would theoretically enable flawless megachips to be developed within a year. Van Meerbergen's group was looking for a suitable project on which to test it. And so cooperation started between the Business Group TV and the product division Semiconductors, on using Natural Motion as a test case.

Van Meerbergen: "It was a great success. With the help of Phideo, a project team under the leadership of Bart de Loore, succeeded in miniaturizing the very large laboratory configuration of Gerard de Haan, and that led to the IC which I am holding in my hand. It still turned out to be a chip of considerable size: 97 square millimeters, or almost a square centimeter. It has a clock-speed of 33 MHz, but consumes less than two watts, and is linked to a microprocessor which checks to see what type of signal (film or video) is being received." "Those who hear the figure of 1 million transistors might think to themselves that there are already memory chips in existence with 16 million transistors. That's true. But those are special chips with only a single function. Melzonic combines many functions. If you were to make a comparison within the same class, this design is at the limit of what can be achieved with today's state-of-the-art. It was possible for us to do it because at Philips Research we had the specific design method know-how in-house."  

"Maybe I can explain it better in terms of the power of the Melzonic. The chip performs 9 billion computations per second. Now, if you take a Pentium processor with a clock speed of 100 MHz, which is probably the best computer chip available today for a PC, you would need at least 100 Pentiums to achieve the same result. Therefore, to equal the processing power of the Melzonic chip, you would have to buy 100 PCs. Simply unimaginable!"  

"We will have to work very hard in the future in order to achieve a similar result." The next generation of chips with 10 million transistors is just around the corner, but there won't be the luxury of being able to take ten years (ten times one year) to produce them. New and more advanced tools are necessary to speed things up. Thanks to the experience we have gained with Melzonic, it won't be an impossible task to create these new tools."