Fundamental shifts in EDA

I envisage two fundamental shifts in EDA. The first is the evolution of integrated EDA tool as distinguished from the traditional EDA point tool. Deeper into nanometrics, problems are becoming interdependent. For instance, physical layout problems become linked, with power impacting timing and area, and all three becoming important factors affecting yield. During chip evolution through various nodes, yield was a function of random defects whose probability was higher over wider areas. As the chip became smaller, random defects became smaller, and yield improved. However, at 90nm yield deteriorated, and it continues to deteriorate further at lower nodes, with optimization problems. Designers feel that a concurrent solution, rather than separate solutions, can alone provide optimization of functions that can give proper yield. Such a solution connects design and manufacturing.

With 65nm round the corner, this fundamental shift should come about soon. The move towards complete solutions should see more acquisitions, as EDA companies will need to acquire one another to get new capabilities. The EDA industry has had more or less flat revenues for some years. I don’t think the change will make any significant impact on the industry revenues; but I do see a consolidation process that will reduce the number of EDA companies. For one thing, EDA has always been through consolidations perhaps more than any other segment of the electronics industry; for instance, Cadence alone has acquired some 50 companies in the last 15 years. The need for an integrated tool flow to enable offer complete solutions is likely to aggregate the consolidation move more than ever before. There will be fewer small companies, and the domination of the three major companies, who already command over 75 percent of the market, will grow.

The second fundamental shift pertains to analog EDA. EDA companies have ignored the analog potential. Analog circuit designs, with their multiple interdependencies for place and route, simulation, etc., have always been more difficult to design, in spite of the fact that the layout of analog circuitry has not changed much over the last three decades. Experts believed that a breakthrough in AI will solve the automated layout problem. This belief for long provided a basis for future EDA research. For decades AI researchers worked on the assumption that the human brain is digital, with neurons corresponding to binary switches. This assumption now seems unsustainable. Electrochemical properties of the neuron show that it resembles an operational amplifier more than a logic gate. This means that the human brain is more analog than digital.



What are the implications for EDA? Hitherto EDA companies have been using digital computers to solve analog problems. But now more and more EDA designers are veering to the viewpoint that EDA companies should investigate building analog computers to solve the problems inherent in analog circuitry.

It seems the use of analog circuitry to design products has already begun. The consumer electronics product Robosapiens toy has been developed using analog circuitry. Its developer Mark Tilden says that many of the competencies difficult in digital are automatic in analog. I think this is natural. The analog prototype should automatically adapt to the real world without the difficulties of digital computing to interface with the real world.

Will analog-centric design processes represent a new design methodology? Will analog computers evolve to design and test analog circuitry? And will the evolution of the analog computer finally lead to the cherished goal of AI, which has deluded designers for half a century? The first two questions are more easily answered. Analog computers, neglected for decades, are poised for a return. Before digital stormed electronics, analog computers were used for sophisticated tasks, including differential analysis, finance and desktop. However, long neglect was a setback to their evolution. The new design paradigms opening up have given them a fillip. EDA will go beyond digital into analog.

I hesitate to answer the third question. Designers say that AI would emerge when computers reach 100 billion gates, which is the number of neurons in a human brain. I wonder whether we have enough understanding of the functioning of the human brain to consider duplicating it. I shall leave the question unanswered.

You can reach Kirtimaya Varma at
kirti.varma@rbi-asia.com