DeepSim, Inc.

TLDR: AI-accelerated 3D physics simulator for engineers

DeepSim makes it easy for engineers to run 3D physics simulations fast. This means rapid design iterations for better products and real-world monitoring for better decisions.

We built a new physics simulation platform that leverages AI physics models for very fast, ultra-scalable physics simulations.

—

Hi everyone! We are Connor, Alex, and Chuck, and we love physics simulations. While working on our PhDs at Stanford, we saw how physics simulators needed a refresh to solve modern problems. That’s why we started DeepSim.

The Problem

Running 3D physics simulations is difficult, expensive, and slow.

1. Setup: Experienced engineers still spend countless hours meshing their designs before even running a single simulation.
2. Run: Many simulations take hours to even weeks on expensive compute systems. Some are even intractable and require highly trained engineers to simplify the problems at a significant cost of accuracy.
3. Attention: Each step requires constant babysitting as simulators are fragile and can fail at any time.

🧑‍🔧️ Engineers hate the tedious simulation setup and wasted time due to failed simulations.
👔 Managers hate that engineers can’t go faster and use tons of cloud credits.
💰 CEOs/VPs hate that engineering delays stall a product launch.

The Solution

Leverage AI to simplify, speed up, and scale up 3D physics simulations.

We are building a new AI physics simulation platform, automating away the physics setup and simulating designs 1000X faster with no drop in accuracy.

Our platform uses lightweight, easy-to-train AI models combined with a powerful GPU-accelerated solver to deliver solutions quickly and efficiently. This approach eliminates the bulkiness of traditional finite element method (FEM) tools and overcomes the inflexibility of other AI physics simulators.

At DeepSim, we have already worked with semiconductor companies to build the most powerful thermal simulator for circuit design, running thermal simulations with billions of nodes in minutes on a single GPU, which is 1000X more detail than a commercial tool can deliver using 50+ CPU cores in the same amount of time.

With DeepSim's simulator, we are able to model extremely complex geometries with length scales spanning 6 orders of magnitude, intractable to solve with FEM tools. Below is a thermal simulation of an integrated circuit, complete with a heat sink and airflow, that only took 10 minutes to run. We can resolve hot spots in individual transistors (~10 nm in size) within a whole chip (1 cm).

Use Cases

This is a breakthrough in 3D physics simulations that empowers engineers in the same way FEM did 50 years ago. We’re excited to see where engineers take our tool!

Some initial use cases are:

• Speeding up the design of complex systems with faster simulations
• Utilizing design automation to speed up development
• Developing digital twins for system operation

The Team

Chuck, Alex, and Connor founded DeepSim to solve the frustrating parts of their PhDs.

The Ask

We would love to talk with engineers that…

• design and operate battery packs for EVs or grid-scale energy storage
• design and deploy satellites
• work on thermal management of semiconductors, from the chip scale to data centers
• have any complaints about slow or frustrating simulations

Contact us at founders@deepsim.io!