. 01 HOW IT WORKS
. 02 About DeepPCB
ADVANCED AI CAPABILITIES
Our latest release covers the following features:
- Routing for Printed Circuit Boards of up to 300 airwires, 4 layers and multi-planes is currently available.
- DeepPCB uses industrial-standard DSN design file format as input. The results are returned to the board editor via SES session files as well as RTE files. All main PCB EDA tools support the Specctra format.
- All our solved boards are Design Rules Checked (DRC clean).
- DeepPCB™ leverages innovative InstaDeep AI technology combined with powerful GPUs and Cloud infrastructure to offer the first pure AI PCB routing engine.
- Smart PCB routing with a time budget of 24 hours.
. 03 Pricing
$0 for 4 boards. Then $9.99 per board
Up to 150 airwires Up to 4 layersGet Started
$29.99 per board
From 151 to 300 airwires Up to 4 layersGet Started
. 04 YOUR QUESTIONS ANSWERED
Is DeepPCB™ fully-automated?
Yes. Automation was possible thanks to Reinforcement Learning (RL), a subfield of Machine Learning (ML). RL is particularly suited for decision-making problems: board games such as Chess or Go, logistics, mobility, or PCB routing. At InstaDeep, we deploy RL systems in the real world and work closely with hardware partners such as Nvidia and Intel. After a rigorous technical evaluation, Nvidia upgraded us to preferred partner, the highest status possible. On top of that, we are part of Intel's AI Builder Program. Users can access the intermediary routing solutions. The user sees how the AI system progressively increases the quality of its routing as it finds out how to solve the user's board.
This is an unsolved NP-hard problem. How much of DeepPCB™ is actually AI? Does InstaDeep have a track record in this domain?
InstaDeep has built credibility in AI circles by innovating in Machine Learning (ML) and Reinforcement Learning (RL), specifically on how to find good solutions to NP-Hard problems. InstaDeep has a track record of publishing cutting-edge AI research in that specific domain at top AI conferences such as NeurIPS, ICML, and ICLR (See our papers). InstaDeep also has expertise in productizing ML and RL research. DeepPCB™ is an example but one can also mention DeepPack™ or InstaDeep's long-term collaboration with Deutsche Bahn, which is on its way to revolutionizing the railway industry in Germany, using Deep Reinforcement Learning (See more).
What are you doing with my data? Can my competitors access it or benefit from it?
InstaDeep will never share or sell any of your data or intelligence to any outside entity. Our sole goal is to accelerate your PCB product development cycle.
Why does it take 24 hours?
Small boards are usually fully routed in minutes. Routing takes more time for bigger boards. In complex combinatorial optimization problems, the search space for solutions is typically very large. To get results, DeepPCB™ needs to spend time finding out how to solve your board. The learning process is the fundamental difference between our system and others. That is why DeepPCB™ is an innovative platform. While 24 hours might appear long and inefficient, it gives the algorithm enough time to find optimized solutions that PCB designers can directly use in their workflows. To illustrate the automated learning taking place, we now release snapshots of the intermediary solutions the system generates during the 24h period. Since we are continuously improving DeepPCB™, future versions will be even faster to route your board and will produce better solutions. Finally, note that as a user, you can decide to stop the run after less than 24 hours if you are satisfied with the solution.
Why would DeepPCB™ be useful if the autorouters that have been around for ages only have limited utility?
This is exactly the problem that DeepPCB™ is addressing. Current auto-routers often do not get the job done because the problem that has all real-life constraints is too difficult for the usual heuristics that current software tools rely on. As the routing problem carries heavy constraints, optimization solvers cannot find solutions in reasonable time. Finally, DeepPCB™ leverages past experience through learning, as opposed to heuristics or solvers who need to start from scratch for each new instance that has to be solved.
Routing at the end of the day is limited. Another difficult part of PCB design is placement. Is anything planned for placement?
Indeed! We are also working on PCB placement for DeepPCB™. Depending on our early customers' interest, we will tweak our roadmap to put more weight on developing placement.
I don't find DeepPCB™ good enough and/or I feel like it misses some key features
DeepPCB™ is a continuously evolving product. Even though we have for now some limitations on the type of boards we currently support, our goal is to hear from you to iterate and improve the product quickly. What matters to us is to work hard and make progress on this exciting problem. We are committed to prioritize the features that our users need and will believe that it will help differentiate our product from the competition. If you have comments or would like us to improve or integrate specific features, please do not hesitate to reach out at email@example.com. We would love to hear from you!
What tools does DeepPCB support?
We support all PCB EDA tools that are compatible with Specctra. In particular, we support OrCAD, Allegro, PADS, Zuken CR-3000/CR-5000, Altium, EAGLE, KiCAD, EasyEDA, etc.
Can DeepPCB continue routing from a partially-routed board?
The best way to use DeepPCB is to manually route sensitive parts of the board, and potentially handle nets for which the routing rules aren’t supported by the tool yet, such as differential pairs and wire length matching. You can leave the rest to DeepPCB. When submitting the board, you can easily select the option to protect the existing routing.