Phản động hay đón đầu?

The EDA industry is located at an interesting place, where anticipation and reaction come together. Too much of either one is wasteful, but too little leaves the industry having to deal with unwanted problems. We see this happening in several areas today, and the balance is changing for several reasons.

We normally expect universities to be 100% anticipatory. There is no point in them working on old problems, unless some change in the industry has altered the dynamics of an old problem, making it a new problem. The industry has relied on academia to invent new transistors, investigate new materials, new architectures, and new algorithms for all sorts of problems.

Most semiconductor companies are reactionary. They use existing architectures, methodologies, and tools to create solutions to satisfy market demands. They are risk-averse, and that means inventing as little as possible, only changing and extending the parts of an existing design that are necessary to reach their goals. There are examples where semiconductor companies have developed new markets because they anticipated a demand that was not currently in existence. Those are disruptive products, and they don’t happen that often.

Foundries sit somewhere in the middle. They do not do the initial research, but they have to turn the theoretical into the practical. They react to problems, but then have to be highly creative to find solutions to them. This is often seen within the yield ramp of a new process, which can identify issues that academia was not even aware of — or at least did not understand the implications. It often involves cooperation with EDA companies when existing tools and methodologies have to be extended to handle new constraints, limitations, or additional parameters that have to be considered in physical design.

EDA is rarely anticipatory because there is no value in creating something for which there is no demand. There have been a few times when EDA took ideas from academia and turned them into something that they successfully managed to persuade the industry that it had to adopt. RTL synthesis is the classic example, but there also have been many failures where EDA tried to create products ahead of demand and lost the entire investment.

Many times, I hear about the EDA industry not being innovative because it is an industry dominated by a few big players. When I was in the EDA industry, it was a common mantra that if you couldn’t offer something that was 10X better than existing solutions, there was no point in even trying. That is the threshold before the industry will take notice, and in some cases, even that is not enough.

The other problem with that mantra is that the 10X has to be for every example. There can be no case when it is worse than what currently exists. As soon as a customer finds an example where the new technology is not better, they will not consider it. This is a high bar, and takes something that is revolutionary – but the industry hates revolution. Too much risk.

Still, the balance is changing. EDA relied on abstraction that separated problems into distinct buckets. Functionality used to be separated from timing, which was separated from power, which was independent of reliability and aging. Those abstractions have broken down, and the interplay between all of these factors is requiring EDA to innovate and be anticipatory. I have been discussing several problems with EDA companies recently where there appears to be a lot of uncertainty about the ultimate value of what they are doing. No one EDA company has exactly the same view into the industry, and they see the specific problems that their customers are talking about. As a result, they are each getting a different picture.

This is especially true for 3D packaging. There are many variations in how this is being approached today, and it is not clear what the solutions of the future will look like. Everyone is innovating and trying things out within enclosed ecosystems. Semiconductor, foundry, and EDA are all reacting to issues being found that require brand new solutions. And the pace of this iteration is much faster than academia can deliver. The industry cannot wait for a PhD to develop a new solution, or even a departmental program to address a need. The solutions are required next week or next month. Anything longer can jeopardize the success of a new product introduction.

The entire industry is in a new phase of rapid innovation, sparked by AI, which in turn is demanding new algorithms, new architectures, new manufacturing processes, bigger chips, better ways to deal with power and heat, and completely new business models. It is requiring levels of cooperation never seen before.

I am not certain how long this new phase of innovation will continue, because there has to be some consolidation and standardization in order for the market to grow into its next phase. That would be the emergence of the third-party chiplet market, but that will take time because there are just too many options. Long term, EDA will go back to being reactionary, because EDA companies will not be the ones to determine what a consolidated industry looks like. At that point, all the EDA vendors will converge on what is essentially a common solution set.