Repositioning For A Changing IC Market

Sailesh Chittipeddi, executive vice president at Renesas, sat down with Semiconductor Engineering to talk about how changes in end markets are shifting demand for technology. What follows are excerpts of that conversation.

SE: Renesas has acquired a number of companies over the past several years. What’s the goal?

Chittipeddi: The goal very simply is to create an industry leading solution provider addressing the industrial, IoT, infrastructure and automotive markets. We are strengthening our market leading position in embedded processing with acquisitions that strengthen our presence in four specific domains — power, sensing, connectivity, and actuation. Historically, Renesas had a major focus on the automotive industry even though it did sell microcontrollers into other markets. What we’ve done since 2019 is to reposition the company into two major business units — Automotive, and IoT and Infrastructure Business Unit (IIBU). Today IIBU has become the bigger part of the company, both through organic growth and the acquisition of Intersil, IDT, Dialog, and now Celeno.

SE: No one chip is going to dominate everything in the future, because we’re seeing a lot more heterogeneous designs even from large chipmakers. How does that fit into your strategy?

Chittipeddi: Yes, you are correct. We are seeing a lot of start-up activity in AI, edge, or endpoint AI companies, and several of these are optimized for specific workloads. A big part of it is really about intelligence moving to the edge. But it’s not just about the CPUs from companies like AMD and Intel, or GPUs from NVIDIA, or TPUs from Google and the rest. As processors get more complex, you need more digital multi-phase controllers and smart power stages. You need memory interfaces and timing devices. We see ourselves as an attach opportunity for these core processor providers.

SE: What changes as the edge takes shape, because the idea in the past was to connect the end point device directly to the cloud?

Chittipeddi: There are a whole bunch of things happening. One is the advent of AI and tiny machine learning that goes into MCUs. A certain level of AI functionality can be implemented in MCUs. The next level is to embed a neural processing unit. NPUs are multi-threaded, but the challenge with either multi-core CPUs or NPUs is the added costs. To address that more recently, we’ve announced our Forge FPGA device. MCUs have limitations because they’re typically single-threaded, which means you have to wait for an operation to complete, before it takes on the next operation. The beauty of the FPGA is there are parallel threads for multi-threading. This is a very low-cost approach — less than 50 cents per part— and ultra-low power. It works well for about 5,000 gates with between 1,000 and 2,000 lookup tables.

SE: So where do you see this going? Will chips disaggregate even further into smaller and smaller components? Or will it hit a limit where it’s too difficult to integrate everything?

Chittipeddi: Some level of integration will always be needed, and that’s going to be workload-dependent. The reason it’s disaggregating is that it’s dependent on what you’re trying to do. So for certain kinds of workloads, MCUs probably will be more than adequate. For others, you’re probably going to need an AI-optimized chip. And for voice applications, for example, there might certain specific neural processing chips. And for vision AI, you might need another category.

SE: It’s becoming more difficult to tell what’s an MCU and what’s a different type of processor, particularly as you add off-chip memory and increase the number of bits, and as you integrate them into packages with other types of chips. So what’s the big driver here?

Chittipeddi: From a high-level point of view, it’s maximum computing capability at the lowest power consumed. So, you need to know what kind of computing capabilities you need, and whether it is optimized for the workload. To your point whether it is data-center side or the edge side memory bandwidth is a major constraint, we have with our solutions provided the highest density memory we can on-board, but there can be scenarios where off-chip memory makes more sense. But the ultimate factor is still the lowest power consumption. When it comes to connectivity the question becomes, ‘Do you put the connectivity onboard, or do you leave it outside?’ But how you optimize for power consumption has to be sorted out at a system level. Maybe that means a single chip that does everything, or maybe that’s not ideal. And if a sensor network is going to be used in a remote area, maybe you’ll want to use NB-IoT, while if it’s in your home environment, you’ll choose Wi-Fi or Bluetooth. For industrial, maybe it’s Wi-Fi 6, which is starting to make inroads in this area.

SE: Why do you see that as a game changer for industrial?

Chittipeddi: Wi-Fi 5 had interference effects, but Wi-Fi 6 has several features ideally suited for industrial applications like massive MIMO, scheduling, target wait-times, extended range, spatial reuse, higher network capacity, signal reliability, lower latency, and an ability to mitigate crowded networks. This, coupled with 1,024 QAM features, Downlink and Uplink OFDMA, and additional frequency band capabilities didn’t exist in Wi-Fi 5. We also will be offering Doppler imaging capabilities for our industrial customers.

SE: So you want to connect more, and you also want to do it faster. But you also now have to adjust to the context in which these chips will be used, right?

Chittipeddi: Yes, and it is dependent on the system. The application you’re trying to address is what determines that piece.

SE: One of the big changes here is that there aren’t many billion-unit designs, anymore, with derivative after derivative. So how do you build in enough flexibility to maximize your investment?

Chittipeddi:  We do see continued emphasis on maximizing die-reuse and using multiple packaging and bond-out options with core embedded processing platforms. We also see application-specific interfaces being used for certain cases, such as Dali-3 for lighting, etc. Moving to a slightly different area, place-and-route is getting interesting for some of the tasks, and AI is moving into that area. From a design perspective, what has changed at the most advanced nodes is the parallelism between the hardware and the software development. In the past, it was more sequential. Now, it’s almost mandatory that you do it in a parallel fashion for the more complex chips, and for the MPU you’d better have a level of parallelism or you’re going to be in trouble. It takes too long for the product development cycle, so it just doesn’t work without that. The customer will now wait for the software to come along. Everything has to be ready and turnkey.

SE: So how does all of this begin changing the MCU?

Chittipeddi: Originally, you had all these companies that were doing proprietary cores, where you had the hardware and the software to give the customer the whole complete solution. With Arm we have created an ecosystem where you have a more flexible software package. Lately, we have been making progress in MCUs with our RISC-V products, the initial one which we optimized for motor control applications. Now, we’ve introduced RISC-V for voice applications. And the nice thing about it is we can extend that concept into other areas, such as a RISC-V IoT gateway MPU. The transformation on the MCU and MPU side is journey, which over time will accommodate all three cores — the proprietary internal cores, ARM cores, as well as RISC-V cores based on application and region specific needs.

SE: Are there enough commercially available design tools, or do you need to develop them internally?

Chittipeddi: For first pass, we relied on a third-party RISC-V core. But we are developing the RISC-V core ourselves so we can optimize those for what we need.

SE: And if you have the compilers, then the idea is that you can optimize them very quickly, right?

Chittipeddi: Yes, but we don’t have a choice. There are customers that are always going to want handholding and a black-box model, and if you don’t support them, you’re in trouble. But there are customers in other regions of the world that don’t want to pay the Arm royalty. And then you have the customers in geopolitically sensitive areas that are worried about lack of access to IP type of issues. For them, RISC-V is the only open ecosystem out there. So you have to operate in all of these diverse environments.

SE: Are you finding that you’re selling to different people than in the past, inside of companies?

Chittipeddi: Yes, for a few reasons. One is COVID, which has changed a lot of things. The level of transparency with end customers versus the OEMs has changed. So now we have more visibility into what we can build for them. We’re not dealing with the people in between. Visibility is at a much higher level than it used to be, just by virtue of the fact that they need the capacity and they don’t believe what people in between are telling them. The second aspect is this whole notion of making the user experience a lot better than it has been in the past drives us to reach out to end users more directly. That has become important. Another element during the CV-19 crisis we introduced is a lab in the cloud. So basically, our customers are able to validate their devices on the cloud themselves.

SE: This is a sandbox type of approach?

Chittipeddi: Yes. That is exactly correct in the case of the lab-on-the cloud. We’re also spending a lot more time offering what we call winning combinations and system solutions. Winning combinations can be as simple as a schematic layout that says for this particular system, combined the the microcontroller microprocessor timing device powered by is on a board for this application. A great example of that would be say, a room humidifier, where a simple layout take is the whole way. The other side of this would be a system solution, where you would define a complete robot arm with all devices put together and deliver the software package and how it would connect to something else. In the past, we would go in as Renesas’ power group or MCU group. That’s no longer the case. We’re going in with much more of an integrated strategy for the customer, targeting vertical markets with everything we’ve got.

SE: So as chips disaggregate and become more domain-specific, you’re putting the pieces together, right?

Chittipeddi: Exactly, and that’s where the breadth of the portfolio and the ability to think through the vertical markets that you want to target become far more important. We’ve created more than 200 of these solutions since we started, and there’s a lot more coming every year.

SE: Alongside of this, there’s a big emphasis on reliability in many of these applications. This is especially the case in industrial and automotive, but even smart phones are now supposed to last four years. How do you improve reliability?

Chittipeddi: For the microcontroller and the microprocessor, in the industrial area increasingly there is a need for FuSa (functional safety), which was wasn’t the case before. That’s a very big change versus five years ago. Traditionally, those requirements came from the automotive side.  Customers increasingly expect reliability to built in with the designs and redundancy for mission critical applications. There is a realization across most areas that responding quickly to market needs is essential for survival. Traditional industrial customers were all about 15- to 20-year lifecycles. That’s definitely changing. People now are looking for 7 to 10 years, and they know they probably have to adapt after that into a newer generation of device. Within the industrial ecosystem, you need security that reaches beyond a standard Trust Zone offering with the latest Arm cores. It has to be tamper-resistant and have other security features. We also offer increased software capabilities in our industrial automation MPUs, with features such as PROFINET and TSN (time-sensitive networking) support. These are things that weren’t done before when the industrial customers largely were reliant on ASICs.

SE: Is that 7 to 10 years you talked about because things are moving from mechanical to electrical? Or is it because change is happening so quickly, that they need to stay current?

Chittipeddi: It’s the latter. The changes are happening rapidly. And the notion of power efficiency, power consumption, and new systems is becoming far more important to people. They’re recognizing that sustainability is not just a couple of points for the stock price. It has to be a way of life, and that is driving a change in behavior — especially in the industrial segment.

SE: Is it still all about silicon substrates, or are you seeing more emphasis on materials like SiC and GaN?

Chittipeddi: It depends on the application, the voltage, and the part. For most applications, silicon, BCDMOS, and LDMOS work well. Above 100 volts, then you’re looking at GaN for the power efficiency you need. Silicon carbide is not that far along for the industrial applications just yet (transmission lines are a different aspect), while in automotive it’s much further along because they’re dealing with 1,000 or more volts. In that space, silicon carbide is getting a lot more attention relative to IGBTs.

SE: So given all of these changes, how will you address these spaces? Will there be more acquisitions, or would you rely on organic growth?

Chittipeddi: The external world is analog, and in our case, we had the embedded processor piece. What we lacked were the pieces that linked it to the external world. It is things like sensing, power, connectivity, and actuation. Those four building blocks are essential for Renesas, and we’ve put them together in a holistic fashion. With Intersil, we got the midrange power. With Dialog, we got what we needed for low power and low-power connectivity. The broader connectivity piece we got with Celeno. And then the actuation resided within Renesas itself. So now you have the full circle built around a bull’s-eye chart of the embedded processing, including the analog and mixed signal and power shield, and it allows us to compete more effectively. Prior to this, we had bits and pieces of it. And we also decided to focus on three major verticals — infrastructure, industrial, and IoT. We will always keep an eye out for acquisitions that make strategic sense, but we believe that we have the critical pieces from a hardware perspective to drive growth organically.