Optimizing designs with co-simulation

Improving complex app design with analog and digital methods

BY BHAVESH MISTRY
General Manager
National Instruments Toronto
www.ni.com

Engineers and EDA tool providers have been in a long-term conversation on how to improve design insight and better predict system behavior prior to prototyping. Best-in-class strategies to design faster, better, and more cost efficiently are based on tools and processes that facilitate such vision into behavior earlier in the design flow. For this reason, the EDA industry continues to focus great attention on simulation technology.

Improving complex application design through modern analog and digital co-simulation techniques

When engineers embrace simulation to visualize design behavior, they improve performance before the first physical prototype. Errors are caught and performance optimized in a CAD tool, reducing the number of prototype iterations and speeding development time. Systems, however, are becoming more complex, and engineers are in need of a new breed of simulation technology to tackle such applications.

Simulation in the traditional design flow

Simulation languages have been tailored to specific design worlds. For example, SPICE has been used for low-frequency analog circuits. Event-driven simulation with VHDL and Verilog has been used for embedded devices. Other simulators have been developed for mechanical motion, logic, RF, etc. to predict behavior for various ICs and chips. These tools are powerful in their respective domains, but engineers know that there is an implicit convergence of these analog, digital and mechanical components.

In the example of power applications, analog and digital circuitry will be coupled in the final design, but each circuit is designed separately. This leaves many a crossed finger that the final prototype will function correctly. Looking back on past projects, engineers know that the inherent complexities of analog circuitry, with harmonics and resonance interacting with digital logic, will not have been addressed, and prototype iterations will be necessary as behavioral errors are found.

Adapting simulation to new complex applications

To overcome such issues, system-level solvers based on linear, time-invariant models have been used in lieu of individual simulation tools in the past. Such tools still do not accurately account for all the coupled relationships between the analog and digital worlds thereby again lacking the accuracy needed for precise design. This has driven the need for nonlinear multidomain simulations. By integrating separate “solvers” engineers can “co-simulate” their entire application and improve their system level insight.

In the circuit design world, co-simulation results in performing linear time-domain analyses concurrently between two solvers. Each simulation engine exchanges data with the other so that the effect of each part of the design is accounted for at a holistic view. The key to such an approach comes down to two simulation products communicating their time steps based upon the dynamic changes in the system and negotiating how best to simulate based upon these current conditions.

Co-simulation in the power domain

A great example of the need for co-simulation is in the power domain. Such applications are based upon a closed-loop system with an analog stage (for example, a switch-mode power supply) and a controller (for example, an FPGA). In a traditional approach, each stage is designed independently with SPICE and event-driven simulation, respectively.

The stages meet as a complete physical prototype where current levels, voltages, switching behavior, and controller performance are finally verified, and often found not to meet specifications. This leads to prototype iterations with PCBs being re-spun and FPGA code recompiled. Each PCB can take 3 to 5 days to fabricate illustrating the immediate loss of efficiency.

When both stages are concurrently designed with co-simulation, the controller can be validated as it actually engages with the analog front end. Simulations visualize current spikes, settling time, and other behavior as the various relationship of closed-loop analog and digital circuitry are validated.

With NI LabVIEW software and the NI Multisim SPICE environment, co-simulation displays accurate simulation results that closely correlate to real measurements. Control algorithms are improved before being deployed and prototype iterations are avoided. With the NI solution, the LabVIEW embedded simulation code is also the same code that will be deployed to the FPGA, making the flow from simulation to prototype seamless (and fast). This is possible because each solver has naturally been optimized for its respective domain, and co-simulation leverages this for precise behavior validation.

For EDA tools providers, this co-simulation strategy is by no means a simple technical solution. As we have discussed, different solvers have separate time-based characteristics to calculate and predict system behavior. However, the need for visualization of complex analog, digital, RF etc… relationships is pushing the EDA industry to address how their tools can deal with these applications. In the case of power applications, we are starting to see how two tools can work together to provide the accuracy and precision needed by the engineer. ■