It is that time again, the Comsol Conference Boston took place last week covering various applications from their partners and the new features of version 6.4 coming out soon. Here is a list of the new features:

• Explicit structural dynamics and mechanical contact
• New Granular Flow Module for simulating the movement and interaction of solid grains in bulk processes
• Nvidia Cuda Direct Sparse Solver (cuDSS) for one or more GPUs providing several-fold speedups for both single-physics and multiphysics simulations
• Higher-quality quad-dominant and swept meshing
• Spatially varying transparency
• Array-based plot layouts
• Support for Open AI API-compatible large language models such as GPT-5™, DeepSeek™, -Google Gemini™, and Anthroipc Claude
• More efficient handling of large simulation apps
• Cluster support for models saved in a Model Manager database
• New time-depedent and parameter optimization options
• Export of network parameters for deep neural network surrogate models
• Surrogate model data generation on clusters
• Frequency- and time-dependent uncertainty quantification

I talked with EM expert, Andrew Strikwerda, about the most important features for RF/microwave engineers and said there were 3 major improvements:

• Far-Field simulation in the presence of a substrate materials
• Multi-conductor transmission line simulation
• A new, user friendly interface

Here is a list of all of the RF/microwave features added in 6.4:

Physics Interface: Transmission Line, Parameters

• Transitioned from a complex Multiphysics setup with two interfaces to a simpler and user-friendly configuration with minimal boundary conditions
• Series resistance, series inductance, shunt conductance, and shunt capacitance per unit length, as well as characteristic impedance and propagation constant
• Frequency-domain modeling in 2D

Physics Feature: Periodic Structure

• Designing metamaterial
• Default Periodic Port and Floquet Periodic Condition features

Physics Feature: Impedance Boundary Condition and Transition Boundary Condition 

Lossy boundary features for time-domain analysis

Physics Feature: Far-Field Domain, Inhomogeneous

Far-field radiation in the presence of a substrate

Enhanced Far-Field Functionality

• Functions for optimization
• Variables for circular polarization

Automatic Port Probe 

Monitoring S-parameters while computing

Part Library

Coplanar waveguide (CPW) trace parts for modeling quantum computing devices easily

Material Library

Rogers corp. (substrate), Hymag’in(absorbing), and Inkbit (3D printing mmWave)

I also talked with Daniel Smith, Founder of PHYSAI, about training neural network models on a dataset (running overnight) and then using that model to quickly hone in on how each parameter affects the model with easy to use sliders to adjust the values. If displays the results immediate using the model as an approximation of the final result. Once you are happy with the adjustments, you can run the final simulation overnight to get an accurate result. We hope to get an article on this technique soon.

We look forward to the release of version 6.4 and will cover it once released.