Executive Insight: Satish Bagalkotkar

Semiconductor Engineering sat down with Satish Bagalkotkar, president and CEO of design services company Synapse Design to talk about massive shifts in the semiconductor industry and his vision of how these changes will alter the landscape, from chipmakers to design services to what gets built and how it will get used. What follows are excerpts of that interview.

SE: What worries you most?

Bagalkotkar: Everyone has been pushing smaller nodes, bigger devices. Now we have gotten to a point where building a semiconductor product by itself is not feasible or cannot be funded. You’re talking about $30 million to build a chip. The volume has to be at a certain point just to break even. The industry cannot support that.

SE: So what’s the solution?

Bagalkotkar: The only way to solve that is to make devices simpler using geometries that are already there. Rather than going 14nm and 10nm and 7nm, people are looking at going backward. Devices need to be more specialized. They should be able to run on a watch battery for two or three years. And the business model has to shift so that instead of buying it, the model is subscription-based. More and more devices can be controlled through software. To make this happen, the devices have to become simpler.

SE: But not all devices become simpler, right? Somewhere you need really powerful processors if you’re going to offload the processing and storage.

Bagalkotkar: That’s correct. On the user side, the devices will just become displays. They may be fancy dumb displays with good network connections. On the cloud side, that’s where you need the 14nm and 10nm chips. There are only going to be a handful of companies building their own fabs. They will partner with the EDA companies to make sure they can build a device and get the performance. But very few companies need to be in that space. On the other side, you can create a 3 billion to 10 billion unit business. That will enable a lot of new startups with a lot of new ideas. And they won’t even have to build a design team internally. They can architect it, partner with a company to do the engineering, and that becomes a good road map. It enables startups to get a product out the door for $8 million to $10 million.

SE: How close are we to that becoming a realistic, mainstream business model?

Bagalkotkar: We’re seeing a few large customers driving in that direction already. The question is when the rest show up. It’s a herd mentality, to some extent. And for some of these companies, they may not even need 28nm or 40nm. They might be okay with 65nm or 130nm. It’s all about what power you need.

SE: Assuming this equation works, what happens in terms of time to market and time to very specific markets.

Bagalkotkar: It shortens quite a bit. The IPs are going to be standard—at least to some extent. We might have to strip out features or performance out of existing devices. That means the time to market will shrink. The question is how much can you strip out for specific markets. If you can strip out 90%, that means your battery life is longer, your cost is lower, and your market gets larger. If you have a device that is $100 with a volume of 20 million or 40 million, a stripped down version that costs $2 can be sold in units of 1 billion in markets such as China and other parts of Asia. If it’s simpler and cheaper, you can deploy more of it.

SE: This is the reverse of the superchip concept. What’s the most efficient way of doing that?

Bagalkotkar: You’re not pushing the technology in terms of more features. It’s all about how compact you can make a product and how you package it. You can’t have a chip that sells for 80 cents and have a package that costs 40 cents. It doesn’t make sense.

SE: This also means reorganizing the entire supply chain. It’s been talked about in terms of just-in-time manufacturing, but not in the high-tech world or for such vertical slices of a market.

Bagalkotkar: Getting a product out has to be simplified. Once you get things rolling and the basic IP and networking are in place, and it meets power and performance and cost, then deploying ‘n’ number of products is very easy. It should be a matter of a few months.

SE: What needs to shift in your business to take advantage of this?

Bagalkotkar: The semiconductor industry already has self-destructed. Our desire to make bigger and bigger and faster and faster chips has driven us to the point where the industry is on its deathbed. Most of the VCs are no longer involved, which means there will be very little innovation. It’s the wrong path. If you say the IP is your architecture, getting a device out the door should be simple. You don’t have to build things anymore. You can have architects define it, and then you work with a services company that is good at what they’re doing. If a services company does this on a daily basis, this is relatively straightforward. There is no learning curve. Everything is in place. The tools and scripts are working. We come in and do what we are good at. We can define which IP makes sense because we’re already working in a number of different sectors, so we can see which IP adds the most value. Then we can work with companies that are good at manufacturing. That will enable a new breed of innovators because the returns will be big. If you can get the product out for $10 million to $15 million—and that includes the chip, the software and the hardware that you can take to market—even if you sell it for $1 you only need to sell 16 million to survive. We’re talking about a market where you can sell hundreds of millions of units. And if your product is not correct, you should be able to change it and bring it back to market within three to four months. That allows more risk-taking because it’s an affordable game.

SE: There is widespread recognition that Moore’s Law as an economic value proposition is winding down. Do you see more companies coming into the semiconductor market as a result of that?

Bagalkotkar: Yes, because once you start creating money, risk-taking happens and innovation comes back. You can go back to doing 20,000 chips or 40,000 chips a year. Once you start creating simpler devices, your risk goes down. You’re going after very targeted elements. The complexity is on the cloud side. That also means the odds of you making a chip that doesn’t function properly are very low. The technology has been around for 10 years.

SE: How does this change the industry landscape?

Bagalkotkar: The FPGA guys make their money on the bigger chips. They will still play a role. The big fabless guys, instead of putting 500 people on a project, will now be able to do 400 projects. And as they make more money, they can make more products. But they also have to do everything faster. Right now if you have $100 million in the bank, you can do a product, but those companies are the slowest movers. Fundamentally that’s like taking an elephant to the horse races. You need real horses for that.

SE: But this also makes it easier to serve new markets more quickly, right?

Bagalkotkar: Absolutely. The market opens by itself. You’re not looking for a new market. You’re looking to make an existing market better and more efficient. The price point is so small and time to market is so short that you can do anything and take the gamble.

SE: Based on this approach, do you see the semiconductor market increasing significantly?

Bagalkotkar: It should explode, and the return on investment should be an order of magnitude higher. Now you’re making money over the long term. If you sell something that costs $1 and it last three years, you’re making money for three years and then you throw it away. That means you’re changing things out every three years, too. You want to get to the point where changing a battery is more expensive.

SE: How much of this gets put into platforms like stacked die?

Bagalkotkar: Those will be for specific markets. If you need a product that is cheap and compact, you have to play with all three dimensions. The device itself should not be visible. It should blend in with the surroundings.

SE: This is the idea behind the Internet of Things, which has been described as a mesh network for data. How do you see this unfolding?

Bagalkotkar: As each element gets connected, everything automatically adapts. When you look at the technology for recognition (and authentication), every time I use my phone I have to put my finger there for two or three seconds until it recognizes me. If you look at what defines me, I have a unique cholesterol level, a unique skin color, a unique voice. There are thousands of unique things about me. If there is a way to recognize that, everybody knows who I am. That makes it impossible to steal my identity. So why do I need to carry a wallet or a credit card? I walk in, I pick up something, and I walk out. That thing I pick up should have a device in there that says I’ve got it. I don’t need to stand in a line to buy it. There’s no reason for anyone to try to steal my identity. It should be impossible. It’s the same when I go to the doctor. My information should already be available because it’s in the cloud. And when I walk into the doctor I don’t need to see what the latest movies are on a device. I only need what’s relevant. When I walk into a theater, I don’t need my medical information. This is all possible using cheaper technology.

SE: So the technology becomes less visible to the user, but the amount of data available is greater?

Bagalkotkar: The technology is invisible. You might need different devices in different markets, but displays are going to be cheap. The hard part is in the back end. The front-end technology is disposable. Every car will have it, every elevator will have it. Doors will get smarter. Cups will get smarter. And you should be able to afford it all very easily.

SE: That also means everything will have to be re-engineered, right?

Bagalkotkar: Yes, completely. You have to start with the thought process that everything is obsolete.

SE: How many other companies will move into the chip creation business?

Bagalkotkar: This industry has to grow 20 times or more. Instead of 40,000 chips per year, why not 200,000? Each one has to be a complete solution. It has to include software. It needs to do specific pattern recognition, which may be a voice. You look at everything you’re trying to do, build a complete solution, and loop it back to the user. Instead of tracking one thing, you’re tracking everything around the world that’s related to you. The data is relevant, but only to you. Everything has a purpose. Then the matrix makes sense. If it doesn’t apply to you, it may only be 50% accurate, which isn’t good enough.