Radar may be best known as a guidance and detection technology used by air traffic control and the military. But with a wave of new demand for smart radar in everything from connected cars to healthcare, a few pioneers are driving the technology to new frontiers. Here, Korea Electronics Technology Institute director Sungho Lee gives us the lowdown on its Radar SoC innovations – and how his team’s ambitions were enabled with Arm’s Cortex-M3 and NIC-400…
Radar is obviously not a new technology. But while it has traditionally been used for military applications, such as missiles, detection systems, and surveillance sensors, that is changing. With the emergence of edge devices, radar sensors have begun appearing in a wide variety of commercial products, from the automotive industry to the IoT.
The basic function of the radar sensor is to detect the range, velocity, and angle of the target. Indeed, most people may still picture radar as a shape moving across a computer screen, accompanied by a ‘pinging’ noise. But with a step-change in the technology, radar can now be used to monitor people’s vital signs remotely. This could be used to detect infants left alone in cars, say, or to track the health of elderly or disabled people. It can recognize people’s hand gestures to operate smart home applications, or combine with AI to guide autonomous vehicles. In this smart sensing, radar is superior to other sensors, and that is driving much of the current demand.
These smart functions demand more intelligent high resolution sensing and signal processing. Radar System on a Chip (SoC) enhances radar’s potential with the latest semiconductor design techniques, integrating high frequency sensors and signal processing engines with various hardware accelerators – all at a tiny scale. Worldwide, there are just a couple of companies that can do this. And no one has done it in Korea. Until now.
Korea Electronics Technology Institute is a non-profit government-sponsored research center. For our initial work on Radar SoC, we have been focusing on object recognition, and the hands-free breath and heart-rate monitoring. We had to develop an SoC with the flexibility to accommodate changes in newly developed algorithms, as the related software may change even after the SoC is built. The added challenge was that sensor SoC is edge computing, which means memory and hardware resource has to be restricted.
A key part of the design was for dedicated hardware-based accelerators to boost the SoC performance. This meant incorporating machine learning, a frequency transformation engine, and highly integrated signal processing units such as AI hardware IP and a radar signal processor. We would need to find the right platform to handle the traffic data between IPs and in/out memory.
At first, we considered using an open-source community core and a self-developed interconnect. But there were several issues. First, our research center does not have enough R&D researchers, so we would need extra time to develop and verify the interconnect, which was not feasible given the time constraints of the project. When you are designing large-scale integration, you have to be able to verify it, even when the project is just for research, not mass production. Second, Radar SoC needs to operate not only at the hardware level but at the software level too. So we knew we’d have to use our hardware resources with well-established development software tools.
We ended up contacting James Lee in Arm Korea. A key member of our team, Jongho Kim, is an expert in digital-signal integrated circuit and SoC design, and he has been using Arm IP since the late 1990s, with the Arm7 core family. As he says, when you design with Arm IP, the company’s design philosophy is always evident. It has evolved clearly and logically from Arm7 to Arm9 and the Arm Cortex® series, so anyone who has designed an SoC using Arm7 can upgrade to a higher CPU design easily. All it takes is a little study. This brings great efficiencies in design. Open-source cores are a different thing. They are developed by various companies, and you do not feel that sense of common connection.
James Lee introduced us to Vanesa Navarro in Arm Research, because our project, which involves grad students and post-doctoral members, is purely for R&D. Once we had demonstrated our use case and how we were working, Vanesa helped us learn how to use the Arm software system. She also showed us around Arm Socrates, which guided us through the selection, configuration and creation of Arm IP. We were very thankful for the help.
We used Arm IP to design our first Radar SoC, two years ago. To provide the CPU performance we needed, we designed circuits around the CPU to be capable of 320MHz, a very high frequency that would enable the software to operate with a variety of algorithms.
Arm Cortex-M3 is the most well-proven 32-bit embedded CPU core for control systems, but it does not usually achieve such a high frequency. But we got there, using the Taiwan Semiconductor Manufacturing Company (TSMC)’s 40nm CMOS process technology. To increase the performance, SRAM was designed to operate with no-wait state on the Advanced High-performance Bus architecture. The throughput between this Bus architecture and the other AXI-based architecture was also adjusted to keep it constant. In this way, we got the increased performance and efficiency we needed. Getting the M3 core to operate at 320MHz was a huge achievement.
Data transfer is always a hot topic in SoC design and was another major factor influencing the design. Here, we optimized the amount of data read and write that occurs at the same time as parallel processing is performed. The CoreLink NIC-400 Network Interconnect provided the AXI-based Bus architecture. Despite being a small edge-device-oriented SoC, this Bus was built with 64-bit operation, to keep processing delays to a minimum and prevent latency. With this approach, we were able to process large amounts of data.
Our relationship with Arm has been very good. Vanesa and James have done so well, and we have never needed to express any concerns to the Arm representatives. In fact, I simply thought of Arm as part of our team.
There were, however, times when I could have asked for more support – with the TSMC the process design kit, for example, or the SRAM generator. But we were able to resolve those issues through our TSMC layout design house and dealer, so I never asked for help from Arm. I would like to do more of that next time.
The first version of our Radar SoC worked out well, and we came out with the second Radar SoC in the summer of 2021. We’re planning to revise and update it again next year: we are interested in finding new permutations, identifying SoC systems that require more sensitive and smart sensing. They will need higher-rate resolution sensing and signal processing techniques. Image radar, video-rate synthetic aperture radar, and nemo-signal processing are all promising candidates. In the future, we will be looking to integrate AI technology with radar sensors, so users should be able to figure out what an object is, even without cameras.
We are standing on the starting line of these new permutations now, but we are planning to explore more, and Arm IP can continue to play an important role in helping us. Arm has the best IP out there. We have already proven that Radar SoC can be realized with M3 and NIC-400, which are the most accessible examples of Arm IP. However, Arm has other, higher performance IP, such as the Cortex-A series, and the mesh type bus system. By combining these different tools, we expect to achieve even better performance.
According to various market analyses, the radar sensor market is expected to grow at around 18% per year. While we are purely a research-based institute, private companies are paying attention to our Radar SoC technology, so it can be commercialized. It’s already in demand from several companies, who are watching how our work is progressing, and calling me semi-regularly to ask how we're getting on. By showing them our Radar SoC with Arm IP, the interest in the commercial possibilities is only going to increase.
Sungho Lee is Director of the Convergence Signal SoC Research Center at the Korea Electronics Technology Institute.
Arm makes a wide range of IP and tools available at no charge for academic research use. To find out more about the IP available, please visit our website.
Explore Research Enablement