EEVblog 1535 - DeepPCB AI AutoRouting FAIL!

EEVblog
28 Mar 202326:16

TLDRIn this video, the creator revisits his previous comparison of manual PCB routing versus Altium's auto router, this time testing DeepPCB's AI AutoRouter. Despite the AI's promise of advanced routing capabilities, the results were underwhelming, with the AI failing to outperform traditional auto routers. The video demonstrates the AI's inability to understand circuit priorities and the importance of strategic component placement, ultimately concluding that the AI's routing was amateurish and not ready for professional use.

Takeaways

  • 😀 The video is a follow-up to an earlier series where the presenter compared his PCB routing skills with Altium's auto-router.
  • 🔍 The presenter is a former professional PCB design engineer and shares his experience that auto-routers require proper setup and constraints to be effective.
  • 🤖 The video discusses the introduction of an AI auto-router by a company called Deep PCB, which claims to use AI technology for PCB routing.
  • 🚀 The presenter tests the Deep PCB AI auto-router using the same file from his previous video to compare the results with his manual and Altium's auto-routing.
  • 🔄 The AI auto-router is in beta version 1.000 and is claimed to be fully automated with no human in the loop, using distributed machine learning.
  • 🕒 The AI auto-router is expected to complete routing for complex boards in less than 24 hours, which the presenter finds impressively fast.
  • 🔍 The presenter notes the importance of component placement in PCB layout, which is crucial before using an auto-router.
  • 📈 The video shows the AI auto-router's process, including its 'rip up and retry' algorithm, which adjusts the routing when it encounters issues.
  • 🚨 The presenter expresses disappointment with the AI auto-router's results, as it did not prioritize routing or produce a cleaner layout compared to traditional methods.
  • 🔧 The video concludes that the AI auto-router failed to outperform traditional auto-routers and lacked the expected learning capabilities from its multiple routing attempts.
  • 👍 The presenter suggests that taking pride in PCB layout work and using traditional auto-routers with specific constraints can yield better results.

Q & A

  • What was the main purpose of the video?

    -The main purpose of the video was to compare the AI AutoRouter by DeepPCB with manual PCB routing and traditional auto-routing software, specifically on the Nixie 2 bar project.

  • What was the outcome of the previous video series where the presenter compared manual routing skills with Altium's AutoRouter?

    -In the previous video series, the presenter demonstrated that their manual routing skills were superior to Altium's AutoRouter.

  • What is the presenter's opinion on the usefulness of AutoRouters in professional PCB design?

    -The presenter believes that AutoRouters are incredibly useful, especially for complex PCBs, but they require proper setup and constraints to produce good results.

  • What was the presenter's initial expectation of DeepPCB's AI AutoRouter?

    -The presenter expected DeepPCB's AI AutoRouter to potentially offer improvements over traditional auto-routers due to its AI capabilities and the claims of being fully automated and cloud-native.

  • What file format does DeepPCB's AI AutoRouter support for the PCB design?

    -DeepPCB's AI AutoRouter supports the DOT DSN file format, which is an OrCAD Spectra file format.

  • What was the presenter's experience with the AI AutoRouter's interface?

    -The presenter found the AI AutoRouter's interface to be good, allowing them to observe the routing process live and choose from different routing solutions.

  • What issues did the presenter identify with the AI AutoRouter's routing results?

    -The presenter identified several issues, including poor prioritization of traces, lack of understanding of the circuit's function, and a failure to complete some of the routes, resulting in an unprofessional and amateurish outcome.

  • How did the presenter summarize the performance of DeepPCB's AI AutoRouter compared to traditional auto-routers?

    -The presenter summarized that DeepPCB's AI AutoRouter performed no better than traditional auto-routers and lacked the expected AI learning capabilities, resulting in a 'complete and epic fail'.

  • What advice does the presenter give to PCB designers regarding the use of AI AutoRouters?

    -The presenter advises PCB designers to take pride in their work, use traditional auto-routers with specific requirements, and manually route boards for better results.

  • What was the final verdict of the video regarding DeepPCB's AI AutoRouter?

    -The final verdict was that DeepPCB's AI AutoRouter failed to meet expectations, did not demonstrate the advantages of AI, and was not recommended for use over traditional auto-routers.

Outlines

00:00

😀 PCB Routing Skills vs. Auto Router

The speaker reflects on a previous video where they compared their manual PCB routing skills with Altium's auto router on a Nixie tube project. They emphasize that while auto routing can be useful, especially for complex PCBs, it often requires a significant amount of time to set up properly with the right constraints and parameters. The speaker's manual routing was superior to the auto router in their past comparison. They also discuss the emergence of AI in auto routing with the introduction of Deep PCB, an AI-powered router that claims to design complex boards in less than 24 hours.

05:00

🤖 Deep PCB's AI Auto Router Experience

The speaker shares their experience with Deep PCB's AI auto router, which is in beta. They describe the process of uploading a DSN PCB file and observing the AI's routing process in real-time. The speaker notes that the AI does not seem to understand the importance of certain constraints, such as high voltage clearance, and does not appear to prioritize routing in the same way a human designer would. Despite the AI's attempts to 'rip up and retry' suboptimal routes, the speaker is critical of the results, finding them to be chaotic and unprofessional compared to their own manual routing.

10:01

🔄 AI Routing's Limitations and Interface Review

The speaker continues to discuss the limitations of the AI routing system, noting that it failed to complete all connections and seemed to get stuck in a loop without improving the routing. They also comment on the interface of the AI routing platform, appreciating the ability to view live updates and choose between different routing solutions, although they express disappointment that the AI did not learn from its iterations. The speaker concludes that the AI's performance was not better than traditional auto routers and that it lacked the expected intelligence.

15:02

🛠️ Traditional PCB Routing Techniques

The speaker contrasts the AI's routing attempts with traditional PCB routing techniques, highlighting the importance of prioritizing certain traces and grouping components together for efficient routing. They point out that a human designer would understand the need to avoid unnecessary complexity and拥挤 traces, ensuring a clean and functional layout. The speaker also discusses the importance of ground planes and power routing, which the AI seemed to overlook.

20:03

🚫 Disappointment with Deep PCB's AI Performance

The speaker expresses their disappointment with Deep PCB's AI performance, noting that it did not live up to expectations for a machine learning system. They compare the AI's routing to amateur work, pointing out specific instances where the AI made poor routing decisions. The speaker concludes that the AI did not demonstrate any clear advantages over traditional auto routers and suggests that taking pride in one's work means manually laying out the PCB for optimal results.

25:04

📣 Final Thoughts and Call to Action

In the final paragraph, the speaker summarizes their overall negative experience with Deep PCB's AI auto router and encourages viewers to give the video a thumbs up if they found it interesting. They invite discussion in the comments and on the EV blog forum and remind viewers of their presence on various alternative channels and platforms, including their own blog and podcast feed.

Mindmap

Keywords

💡PCB routing

PCB routing refers to the process of designing the electrical connections, or traces, on a printed circuit board (PCB). It is a critical part of PCB design that ensures components are electrically connected as per the circuit schematic. In the video, the host compares manual PCB routing skills with automated routing, emphasizing the importance of proper setup and constraints for effective automated routing.

💡Auto router

An auto router is a software tool used in PCB design that automates the process of creating electrical connections on a PCB. It uses algorithms to determine the optimal path for traces based on given constraints. The video discusses the limitations of auto routers and how they require proper setup to achieve satisfactory results, contrasting them with the host's manual routing skills.

💡AI Auto router

AI Auto router denotes a routing tool that incorporates artificial intelligence to improve the PCB routing process. The video explores a product from Deep PCB, which claims to use AI technology for routing. However, the host expresses skepticism about its effectiveness after testing it against traditional methods.

💡Nixie tube project

The Nixie tube project mentioned in the video is a specific PCB design challenge that the host had previously tackled. It involves designing circuits for Nixie tubes, which are a type of gas-discharge display. The project serves as a test case for comparing manual and automated routing methods.

💡Placement

In PCB design, placement refers to the strategic positioning of electronic components on the board. It is considered a crucial step, often quoted as 'PCB layout is 90% placement', because proper placement can simplify routing and affect the performance of the circuit. The video emphasizes the importance of placement before routing, especially when using auto routers.

💡Constraints

Constraints in the context of PCB routing are the rules and limitations set to guide the auto router in creating the electrical connections. They can include trace width, spacing, and avoidance of certain areas. The script illustrates the necessity of setting constraints to direct the auto router effectively.

💡DRC checked

DRC stands for Design Rule Check, which is a process used to verify that a PCB design adheres to a set of predefined manufacturing and design rules. In the video, the host mentions that the AI Auto router's results are DRC checked, ensuring that the routed design meets industry standards.

💡Reinforcement learning

Reinforcement learning is a type of machine learning where an agent learns to make decisions by performing actions in an environment to achieve a goal. The video script mentions reinforcement learning as part of the technology behind the AI Auto router, suggesting that the system learns from its routing attempts to improve over time.

💡Rip up and retry

Rip up and retry is an industry term used to describe a process where an auto router removes and re-routes traces that do not meet the design requirements. The video shows the AI Auto router employing this method, but the host criticizes its execution as ineffective.

💡Cloud infrastructure

Cloud infrastructure refers to the resources and services provided over the internet, which can be used for computing, storage, and other functions. The script mentions that the AI Auto router leverages cloud infrastructure, along with powerful GPUs, to offer a supposedly advanced routing solution.

💡Double-sided PCB

A double-sided PCB is a type of printed circuit board that has components and traces on both sides. The video uses a double-sided board as an example to demonstrate the routing process and to compare the effectiveness of manual routing versus the AI Auto router.

Highlights

EEVblog revisits the challenge of PCB routing with AI, comparing it to his previous manual routing skills.

AI AutoRouting by DeepPCB is tested against traditional methods and the presenter's own routing.

The presenter explains the importance of proper setup and constraints in AutoRouters.

DeepPCB's AI AutoRouter is still in beta, with some users reporting good results.

The presenter notes the limitations of AI in understanding schematic information and circuit behavior.

DeepPCB's service is free for two-layer boards during the beta phase.

The presenter discusses the importance of placement in PCB layout, emphasizing it as 90% of the work.

Live routing demonstration shows the AI's process, including rip-up and retry algorithms.

The AI struggles with high-voltage trace routing, lacking human intuition for circuit priority.

The presenter critiques the AI's routing choices, pointing out inefficient and amateurish results.

DeepPCB's AI fails to complete all routing within the expected 24-hour timeframe.

Comparison between AI and manual routing shows a clear preference for human expertise.

The presenter suggests using AI for less critical routing tasks, while prioritizing human routing for important traces.

Final judgment on the AI's performance is negative, with the presenter deeming it a failure.

The video concludes with a recommendation to stick with traditional AutoRouters and manual routing for quality.

The presenter expresses disappointment in the AI's lack of learning and poor routing decisions.

A call to action for viewers to engage with the content, share their thoughts, and follow on various platforms.