Technical Articles and Newsletters

Improving the Efficiency of IC Development with Model-Based Design

By Kyoji Marumoto and Hiroshi Nishide, ROHM Co. Ltd.

In response to increased competitive pressure, integrated circuit (IC) manufacturers are shortening delivery schedules even as designs grow more complex and customer expectations for quality and performance increase. Many manufacturers are finding that traditional design approaches, in which teams perform document-based verification of the specification and produce multiple prototypes before the final production version, are now too slow to keep up with the industry’s current pace.

At ROHM, we’ve incorporated Model-Based Design into our IC development process for motor control applications, sensor applications, and power supply systems. Modeling and simulating mixed-signal IC designs, plants, and microelectromechanical systems (MEMS) in Simulink® have enabled product teams to verify design specifications at a high level before proceeding with circuit-level design. This approach reduces rework, development time, and the number of prototypes while increasing overall design quality. For example, by automatically generating Verilog® code from models that we’ve created and verified in Simulink, we can reduce the verification time from a month to a few days. This not only improves development efficiency but also improves quality by reducing the number of implementation bugs to zero. With Model-Based Design, we can prototype a product—which has already verified the model-level specification and confirmed circuit-level functions and characteristics meet the design specifications—once, instead of three or four times, and go directly from prototype to mass production.

In this article, motor and sensor fields are presented.

Model-Based Design for Motor Control ICs

When developing ICs for motor control applications, our teams begin the design process by modeling the motor to be controlled. We model the mechanical and electrical characteristics of the motor in Simulink using equations of motion and the voltage equation, and then use MATLAB® to fit the parameters of this model based on measured values from an actual motor. Depending on the motor model designed by our teams, we can also incorporate the effects of magnetic saturation by the inductive sensing controls and the effects of wow and flutter due to shaft misalignment. As part of the plant model, we include a model of the motor’s driver transistors that we create with Simscape™ (Figure 1). This driver model enables us to analyze transient characteristics; for example, current oscillation at the start of pulse width modulation that’s caused by parasitic capacitance in the motor winding.

Figure 1. Motor control and plant models in Simulink.

Figure 1. Motor control and plant models in Simulink.

We model the motor controller in Simulink as well, and then run system-level simulations with the controller and plant together to check the speed, position, and rise of control functions of the design. After verifying the controller design in this way, we use Fixed-Point Designer™ to convert the control algorithms to fixed point. We then generate synthesizable Verilog RTL from the model with HDL Coder™, accelerating implementation and eliminating the risk of introducing coding errors that we previously encountered with hand coding.

Developing MEMS Devices with DPI-C Model Generation

For projects involving MEMS sensors and associated sensor ICs, we use a development process much like the one we use for motor control ICs. Instead of performing tests to characterize a motor, however, we use 3D electromagnetic analysis and structural analysis tools to characterize the MEMS device, and then fit the parameters identified through this process to a Simulink model of the device. Alternatively, we perform transfer function identification and multiple regression approximation in MATLAB, then use the transfer function as a model of the device.

We create a Simulink model of the sensor IC, which, much like the motor controller model, serves as an executable specification of the design. Through system-level simulations in Simulink, we verify this specification early before refining the design in the Cadence® Virtuoso® platform.

In our MEMS design workflow, we can perform an additional verification step that is not part of our motor workflow. Specifically, we use HDL Verifier™ with Embedded Coder® to generate a SystemVerilog DPI-C model from our Simulink MEMS device model (Figure 2). We then use this SystemVerilog model within the Cadence environment to fully validate our IC design—including amplifiers, analog-to-digital converters, and digital processing logic—as we continue to elaborate it prior to sign-off verification. This technique not only increases development efficiency but also helps with the assurance of design quality because we have consistent verification of the design, first in Simulink and then in Cadence Virtuoso.

Figure 2. Workflow diagram for DPI-C model generation.

Figure 2. Workflow diagram for DPI-C model generation.

FPGA-in-the-Loop Customer Evaluations

Many of our customers find the ability to evaluate a ROHM product under development is a significant advantage in their own development processes. For these customers, we generate HDL code from our Simulink IC model using HDL Coder and deploy it to an FPGA evaluation board. The customers can then use the board in evaluations of their hardware designs. Alternatively, customers can use HDL Verifier to perform FPGA-in-the-loop simulations with their own system-level Simulink models for transient analysis and design optimization. With both approaches our sensitive IP is protected, since we are sharing only the FPGA implementation, not our source design assets.

Establishing a Model-Based Design Group at ROHM

To help product teams across ROHM adopt Model-Based Design, we formed the Model-Based Design Group, a team of engineers with extensive design experience. This group develops assets that make it easy for teams to apply modeling, simulation, and code generation in Simulink as part of a top-down IC design workflow. Assets include model templates, documentation, and tools (for example, tools for parameter extraction), as well as a technical guide for motor models, MEMS models, and SystemVerilog DPI-C generation.

The group also shares modeling techniques and conducts internal briefing and training sessions to help teams get up to speed quickly. While the group initially targeted ROHM teams based locally in Japan, it is now helping ROHM’s overseas design centers form teams that specialize in Model-Based Design projects.

Many ROHM teams have readily adopted Model-Based Design, though a few have been reluctant because they haven’t established a Model-Based Design environment for their field. For these latter teams, the Model-Based Design Group takes time to demonstrate the benefits of the approach and the advantages realized by teams already using it. More recently, we’ve set up working groups for sensor IC and motor IC development using Simulink. Engineers across ROHM join these groups to share technical information and learn more about topics relevant to many teams, including how to model MOSFET drivers in Simscape, how to create highly accurate MEMS models, and how to identify the frequency response of existing circuits.

Expanding the Use of Model-Based Design Across ROHM

The number of teams using Model-Based Design within our division is steadily increasing. In addition, we are starting to see Model-Based Design applied across business units throughout the company, including units responsible for the development and manufacturing of silicon carbide (SiC) and insulated-gate bipolar transistor (IGBT) products. Recently, we’ve also seen increased demand from automotive customers for Model-Based Design. ROHM is now well-positioned to meet this demand.

About the Authors

Kyoji Marumoto and Hiroshi Nishide lead the Model-Based Design Group at ROHM Co. Ltd. Their efforts to champion Model-Based Design use across the ROHM organization have contributed to improving HDL code generation and optimizations for motor, sensor, and power IC design.

Published 2022