Better, Not Best

There is little point creating the best possible device when slightly better is usually good enough.


The semiconductor industry has been lulled into a particular way of thinking by Moore’s Law. It is like the age-old joke — you don’t have to outrun a bear, you only have to be faster than your companion. The same has held true for designs. There is little to no point being the best, you only have to be good enough to be better than the competition. That sets the bar.

Best is also relative. While a company may want to buy the lowest power solution, they do not know how low-power it could be. They only can look at it comparatively. Over time, a vendor may identify that a certain class of customer is willing to pay more for a lower-power solution and will invest to make that possible. They move the bar for the other players.

Moore’s Law made the bar a moving target. Every 18 months a new technology node would come along and provide lower power, or higher performance, or additional capability. Often the developer didn’t need to do anything to get those advantages. This muddied the waters because every new design appeared so good in comparison to the previous node that few questioned if the gains provided were even as much as should have been expected.

As Moore’s Law slows, or at a minimum becomes uneconomic for many, companies are having to re-evaluate priorities. Some are looking to advanced packaging to boost performance or lower power, but the costs of those are also prohibitive to many. Those costs may come down over time, but that does little to help if you need to make improvements now.

There is no getting away from it. The only way to improve a product is to improve its design. This is going to change the underlying structure of development teams. Stitch and ship will no longer work. That is a commodity design. Teams have to answer the question, “How do I make my design better?”

This is one of the big drivers for RISC-V implementations. It has become possible for a team to look at the processor and ask, “What capabilities does it have that I do not need?” They don’t care about being RISC-V compliant if they are the only ones who see and program the processor. Do they need that mode of operation, that instruction, that level of performance? If not, they can probably find significant savings.

One of the other changes this will bring is that companies will need to do more upfront design and analysis. They will need to understand the use cases and how those translate into the capabilities of each of the components of the system. Design will have to become more top-down rather than the bottom-up integration, which has been the norm.

I believe that these changes will be good for the whole industry. It will allow companies to differentiate their designs more. We will see a much broader range of offerings, each optimized in different ways. Commoditization will impact only the pieces that companies choose not to invest in – the way it should be. Over time, that will lower the cost of those while allowing premium pricing for products that can differentiate. The market will be opened for companies that want to try different approaches.

It will be good for the EDA industry because more time will be spent in design. It will allow EDA to differentiate products that have become somewhat commoditized too. If they can help their customers create better designs, faster, with more optimizations — then they will be able to command a higher price for the tool. It will mean that some of the older tools may see increased investment or innovation. Today, most of the improvements only happen because the latest nodes need them. EDA, just like design companies, have never bothered to fully optimize things that relate to older nodes.

And, of course, it will be better for the consumer. There will be an incentive for companies to help us understand the differentiations of their offerings. Today I really cannot make buying decisions based on the power devices will consume or how secure they are. I care about these quite a bit more than performance most of the time. Performance is adequate to my needs, and unless a new capability comes along to make me reconsider that, my focus will be on other factors.

You still have to be better than your companion to save yourself from the bear, but now you have multiple ways you can do that. Running may not be the only option available.

Leave a Reply

(Note: This name will be displayed publicly)