From Power, Motors, and Sensors to Edge AI: ADI Portfolio Answers the Reliability Requirements of Industrial Robots Interview with Park Sang-kwon, FAE at Marknica Korea

Amidst this trend, the first question robot developers face is surprisingly simple: what components should be used to build the robot? From stable power supply and precise motor control to unwavering posture recognition and the growing importance of Edge AI, the ICs required for a single robot are numerous and demanding.

We met with Sang-Kwon Park, a Technical Support Officer at Marknica Korea, the official distributor of Analog Devices (ADI) in Korea, to hear in advance about the message to be covered in the upcoming ADI Robotics Integrated Solutions Webinar. As a General FAE, Park supports various fields, but he cited robotics as the area with the fastest growing number of inquiries recently.

Entry of new robot startups and inquiries flocking to power supplies and motors

According to Park FAE, the entry of new robot companies into the Korean market has increased noticeably over the past one to two years. He explained that this is not unrelated to the increased government support.

As interest in robots grows in the domestic market, the number of newly established companies is increasing, and demand is also on the rise. There is also a lot of government support.

The ADI product range handled by FAE Park is not limited to single dedicated robotics chips, but encompasses various ICs that constitute robot systems, such as ADCs, OP-Amps, power supplies, and motor drivers. Among these, the inquiries recently received by Marknica Korea can be categorized into two main areas: power design and noise mitigation, and motor control. Park’s diagnosis is that the harsh environments industrial robots face—such as high heat, vibration, smoke, and humidity—and the reproducibility required to repeat the same task tens of thousands of times, ultimately determine the criteria for component selection. This is also why most of the products recently released by ADI are industrial-grade, guaranteeing operation from -40°C to 125°C.

The structure of this webinar perfectly reflects the flow of this inquiry. It surveys the entire robot system, starting from the power supply and extending to motors, sensors, and Edge AI.

The starting point of all robots, the power chain

The first area that Park FAE deals with is power. The four components—MAX17614 (Protection), LTC7891 (Power Controller), MAXM17572 (µModule), and ADuM362N (Digital Isolator)—form the framework of the robot power chain.

When examining the individual specifications of each product, there are alternatives to ADI. However, a common problem that arises when sourcing components from multiple vendors is that liability becomes unclear when issues such as noise or EMI occur. Power issues that robots must withstand—such as motor inrush, back EMF, and communication line noise—are interconnected, making them difficult to resolve at the level of a single component. Park FAE explains that this is when the value of consolidating into a single vendor becomes apparent.

We have accumulated extensive experience from years of IC development and know-how on responding to diverse field environments. A major strength is our system in place where ADI headquarters works together to identify and resolve any issues that may arise in the field.

In robots that come into direct contact with humans, such as collaborative robots, galvanic isolation is directly related to safety, going beyond a simple EMC issue. This is because high-voltage noise generated in the motor drive unit can lead to malfunctions if it flows back into the control unit. This is also the reason why iCoupler-based digital isolators like the ADuM362N are replacing optocouplers.

The MAXM17572 µModule is frequently considered for applications with significant space constraints, such as narrow joint modules. This is because the inductor and compensation circuit are integrated within the package, which reduces the number of external components and allows for the compression of both board area and BOM costs.

Motor Control Moving to a Single Chip, and Application Scope

The product highlighted by FAE Park in the motor control section is the TMC6460. ADI acquired Maxim Integrated in 2021, and the motor control IC line from Trinamic, a German company acquired by Maxim prior to that, entered ADI's portfolio through this process. The TMC6460 is also part of this trend.

Park, the FAE, calls this product a "micro module." This means it is a one-chip solution that integrates the motor control block, gate driver, and output FET into a single chip. The greatest value of one-chip integration is development speed. In the past, controllers, gate drivers, and FETs had to be selected separately to match interfaces, and protection circuits had to be built individually; however, with the TMC6460, motors can be driven at the code level by simply inputting parameters into registers. Using the GUI environment and evaluation boards provided by ADI also reduces the time required for prototyping.

However, Park FAE frankly limited the scope of application.

"We are targeting smaller motors, such as those for robot fingers. Motors in the tens of watts range are suitable. Large motors in the tens of kilowatt range require a different solution."

The TMC6460 is suited for applications where precision is critical despite low output, such as robot fingers, grippers, and small joints. It was also explained that a combination of a separate gate driver and an external FET is required for applications demanding outputs of several kW or more, such as industrial robot shoulder and waist joints or AGV wheel motors. It is worth noting that an FAE handling integrated solutions clearly explains the limitations of their own product.

The two axes that make up the motor control chain together are the ADA4255 current measurement amplifier (CSA) and the ADA4571 AMR position sensor. The CSA is the core of current measurement that determines the accuracy of the FOC algorithm, and the AMR position sensor is a method that creates a compromise between the price and size constraints of optical encoders and the precision limitations of Hall sensors.

IMU and accelerometer, look the same but occupy different positions

In the field of posture recognition, IMUs and accelerometers are often confused. While both sensors measure acceleration, they play different roles in actual robot systems. Park, an FAE, explained this division of labor most clearly.

The IMU receives data to maintain the robot's posture. It is located on the head or body to determine how much the robot is tilted and whether it is moving forward or backward. Accelerometer sensors detect minute vibrations or movements, and products like the ADXL382 are often used for sensing at the edges of robot hands.

Industrial IMUs like the ADIS16575 are responsible for the balance and navigation of the entire robot. For an AMR to navigate stably within a store or factory, it must estimate its position and attitude using SLAM algorithms, and the accuracy of this calculation ultimately depends on the bias stability of the IMU. On the other hand, low-noise accelerometer sensors like the ADXL382 detect minute shocks or vibrations occurring at extremities, such as gripper tips. They are used not only for collision detection but also for predictive maintenance (PdM), which accumulates vibration data to detect bearing wear in advance.

Park, the FAE, explained that the burden of data volume is not a cause for concern. He noted that even general MCUs can handle it sufficiently using CAN or SPI interfaces, and the amount is much smaller compared to image sensor data. He added that the "scenes of robots suddenly falling over" occasionally seen in videos are often caused by insufficient software exception handling rather than sensor performance.

Why Consider 3D ToF Instead of LiDAR

In environmental perception for AMRs and collaborative robots, the ADTF3175 3D ToF camera is being discussed as a new option between LiDAR and stereo cameras. The ADTF3175 is a 1-megapixel indirect ToF module that provides ±3mm accuracy and a 75° field of view over a range of 0.4 to 4 meters.

Park FAE summarized the difference from LiDAR as follows.

LiDAR receives data linearly, the processing is complex, and the configuration cost itself is very high. On the other hand, since 3D ToF receives distance information like image data, it can be easily processed without complex algorithms. If a Jetson solution costs several million won, ToF can be implemented at a much lower cost.

While long-distance measurements in the tens of meters range like LiDAR are difficult, most hazards faced by indoor AMRs are within 5 meters. This is where the cost and computational advantages of ToF come into play in recognizing nearby obstacles such as human legs, fallen boxes, carts, and other robots.

Extracting depth from standard cameras requires stereo matching or monocular depth estimation algorithms, and running these in real-time necessitates a separate board equipped with a GPU. In contrast, since the ToF sensor itself outputs distance values, the host processor only needs to make decisions based on the received data. ADI has also released ROS packages and open-source reference designs for the ADTF3175, which is another factor driving adoption, as it eliminates the need for developers to build the system from scratch.

The reason for the increasing trend of bundling ToF for environment perception, IMU for attitude perception, and AMR for motor position measurement into a single ADI is ultimately the consistency of technical support. Park, the FAE, also highlighted this as a core value. He noted that the difference between receiving support from dispersed vendors and receiving support from a single line is most keenly felt during the early stages of development.

MAX78002, a card the market is not yet familiar with

The final topic of the webinar is the MAX78002 AI MCU. Developed by Maxim and joined the ADI portfolio following ADI's acquisition of Maxim, this chip is characterized by the inclusion of an Arm Cortex-M4F CPU, RISC-V cores, and a separate hardware CNN acceleration engine on a single chip. The inference for recognizing objects and people using a camera or detecting voice commands can be completed within the chip.

Its positioning is clearly different from that of high-end AI computing boards like NVIDIA Jetson. While Jetson is a board that handles complex visual recognition, SLAM, and multi-sensor fusion simultaneously, the MAX78002 is an MCU-class solution focused on low-power distributed processing of specific recognition tasks. The fact that inference can be completed without the need for a cloud or GPU makes it attractive for industrial environments. A scenario where vibration data collected by the ADXL382 is inferred on the MAX78002 to implement predictive maintenance is also a combination frequently proposed by ADI.

However, Park FAE frankly admitted that the market awareness of this category itself is still low.

"One regrettable aspect is that many people are not yet familiar with this type of AI MCU because they haven't experienced it. I believe that once they try it out, they will realize how good the product is and use it extensively, but from a market perspective, it is unfortunate that the process leading to that first experience is still lacking."

It is natural for engineers to think of boards like NVIDIA Jetson or Google Coral first when considering AI solutions. However, FAE Park’s market assessment is that the value of the category of ultra-low-power AI MCUs, which complete inference with mJ energy, is difficult to gauge without experiencing it firsthand. This can be interpreted as meaning that he hopes the webinar will serve as a gateway to that market.

The Value of Bundling to a Single Vendor, and Maknica's 3-Tier Support

The final message Park FAE emphasized was that the value of an integrated solution is revealed not during normal times, but when a problem occurs. Marknica Korea's technical support operates in three stages. The process involves the client resolving the issue internally, followed by technical support from Marknica, and finally collaboration with ADI headquarters. Issues that Marknica could not resolve are forwarded to headquarters along with accumulated improvement directions, and the structure is such that headquarters finds a solution by recreating the same environment.

Each individual IC possesses sufficient performance and reliability, and since we provide all necessary references, evaluation is possible at the IC level. We provide step-by-step guidance on issues that may arise when combined, and if actual problems occur in the field, we offer solutions to resolve them in collaboration with our headquarters.

This three-stage support structure is particularly meaningful for robot startups facing manpower shortages. The speed at which component vendors work together to solve problems—such as noise issues occurring just before mass production, insulation problems discovered right before EMI certification, and intermittent motor stoppages in the field—directly impacts the launch schedule. This is why FAE Park emphasized "accumulated experience" and "collaboration with headquarters" throughout the interview.

The upcoming ADI Robotics Integration Solutions webinar will reveal how the 11 ICs discussed in this article are connected in actual block diagrams, as well as which reference designs can be utilized in each area. It is an opportunity worth checking out for engineers who are considering how to design the entire system, from power to Edge AI, rather than just a single part of the robot.