Skip to main content

Machine Learning Circuit Simulation Expedites Workflows

Key Takeaways

  • What are some challenges of machine learning circuit simulation and how can they be circumvented?

  • How does machine learning implementation differ from traditional simulators?

  • The use of simulation in power prediction.

 Sine wave on signal analyzer

Circuit simulators are necessary tools for checking intrinsic signal characteristics

A crucial element of design for feedback and testing is circuit simulation. Although simulation environments cannot capture the entirety of circuit characteristics, they offer some insight into the operation of the board and can be used to get an early glimpse at important resources such as timing diagrams and other signal parameters. Circuit simulation suffers from tradeoffs common to heavy computational tasks–the more accurate the software, the longer it takes a machine to render the information (and vice versa). 

Luckily, machine learning circuit simulation aims to bridge the two extremes by operating abstractly while still providing highly accurate results. As machine learning as a field continues to develop and grow, solutions like those found in circuit simulation can only be expected to become more robust.

Approaching Machine Learning Circuit Simulation Challenges

The time cost of simulation in machine learning is the most significant overhead during training of the model. To combat this, new and more sophisticated models must be employed to more quickly increase the rate of development time. Deep neural networks, among other models, can be used as a predictive tool in machine learning simulations for circuit design.

One solution to reducing the computational load as well as the time is comparative simulation values as opposed to absolute. In the former, two circuits are fed as inputs to the model whereby it only has to choose between the greater of whichever two parameters its measuring. Rather than having to fit each individual parameter, comparative selections greatly reduce the time necessary to dedicate to model training. 


Machine learning circuit simulation also runs into problems during analog circuit design. Due to the presence of parasitics inherent to all components, calculated values pre and post layout can differ significantly. This is not a fault of the model, but rather an inherent design challenge between the two stages of layout. Parasitics cannot be accurately accounted for until after the initial stages of revision are done (but before iterating the model), as the nearby layout of components and traces as well as the shape of the circuit will be the ultimate driver of these circuit contributions. 

Additionally, parasitics have sometimes been handled by a human designer as an estimation than a more rigorous approach. This naturally leads to more difficulty in designing how a machine learning model handles parasitics in its design but also presents an untapped pathway for optimization in circuit simulation. With some small tweaks to design and necessary adaptations required of switching from a human operator to a machine learning model, it is possible to not only cut down on the number of iteration rounds but also increase the quality of the final design product. As is common with many machine learning solutions, the model is less flexible and in need of greater generalization for simulation of a wider variety of circuitry.

Advantages of Machine Learning Over Traditional Methods

Even in its relative infancy, machine learning circuit simulations possess a significant leg-up on traditional simulators in both accuracy and efficiency. A veritable bushel of models has been created and revised in an attempt to find the best solution for both specific and general problems in the machine learning design space. First among these for circuit simulation are support vector machines (SVM), neural networks, and random forest techniques, alongside simulated annealing for general optimization. There are tradeoffs between all three methods that better suit them to particular solutions:

  • Support Vector Machine - Defines a line that best distinguishes between two distinct data classifications. The better the line separates the two data classes, the easier it is for the model to sort.
  • Random Forest - A training method using random decision trees and taking aggregate or average decisions for model training purposes.
  • Neural Network - Data with known input and result values are fed to a system and weight is formed that indicates the strength of the connection.


Vector map representation of neural network

Machine learning models, like neural networks, offer some express advantages in circuit simulations

Power Design’s Role in Circuit Simulations

Determining the power draw of a circuit is crucial to the success of the circuit–power design will heavily influence all stages of design, beginning with the stack up and continuing through layout and routing. To more accurately assuage power demands, simulations can be run at high and low levels of abstraction, with the general tradeoff between these ends being accuracy or speed. Machine learning solutions piggyback off the higher-level abstraction simulations to offer a more robust and expedited analysis to assist in the modern quick-turn environment of PCB design.

One current solution for power design simulation involves using high-level programming language descriptions (typically C/C++) and some hardware information to form the backbone of analysis. Compared to more accurate but computationally rigorous and time-intensive approaches employed at the gate level, the former analysis method is able to reach a similar level of accuracy much more rapidly.

View of power circuitry on motherboard

Accurately predicting power circuitry load reduces design iteration cycles

Machine learning circuit simulation is increasing the flexibility of engineers and layout designers; the full suite of Cadence PCB design and analysis software tools contains both software simulation and machine learning modeling to improve your current and future design workflows. 

Leading electronics providers rely on Cadence products to optimize power, space, and energy needs for a wide variety of market applications. If you’re looking to learn more about our innovative solutions, talk to our team of experts or subscribe to our YouTube channel.