Smart MCU: STM32 Board
- Routed
- 100%
- Run time
- 18m 8s
- Est. by hand
- 8 hrs
- Est. savings
- ~7.7 hrs
Caught the untyped power layers and the missing USB pair, and fixed both before routing.
Inspect the full test reportDEEPPCB FOR ENTERPRISE
Your Cloud. Your Tools. Your Rules.
A learned routing engine, seven years in the making: it takes the routine routing volume, and your engineers keep the critical nets. You get the tool, and you drive it.
Not Whether It’s Good.Whether It’s Good on Boards Like Yours.
Find the boards that resemble your team’s work, in their density and their layers and the interfaces they carry, and read the runs the way your team would read them: how much routed, and how quickly, and what was left at the end for an engineer to finish.
STRF: RF Mixed-Signal board
- Routed
- 100%
- Run time
- 2m 6s
- Est. by hand
- 2-3 hrs
- Est. savings
- ~2-3 hrs
Corrected the USB pair impedance against the real stackup before routing.
Inspect the full test reportMore use-cases coming soon
We're benchmarking more boards continuously.
It Runs Where You Run.
Deployment is a policy decision, not an engineering project: the same engine runs inside whatever boundary your data policy draws.
YOUR TEAMS REACH IT
API
Your pipelines. Submit boards and pull routed results from your own tooling.
NATIVE PLUGINS
Your CAD tools. Route from inside the tools your engineers already run. No export, no context switch. Which tools, by name: the toolchain table below.
SELF-SERVE WEB APP
Zero install. Upload, route, and review in our standalone web-app. Run in minutes.
ONE ENGINE
The same learned router, wherever it lives
IT LIVES WHERE YOUR POLICY SAYS
ON-PREM
Inside your walls, on your hardware. The full engine, deployed in your environment and administered by your people. Board data never crosses the boundary.
PRIVATE CLOUD
Your account, your keys. It could be in your AWS, Azure, or GCP account with your logs and controls, inside the perimeter your security team already governs.
HOSTED CLOUD
Routing in minutes. Upload and route the same day. Data handling per the Terms of Service.
Pick the row your policy requires; the way your engineers work doesn't change.
AND WHEREVER IT RUNS, WHAT'S YOURS STAYS YOURS
IP Protection, In Writing.
Your boards are not training data. The engine does not learn from the files you upload, and nothing about your designs ever improves anyone else’s results.
Your boards are yours.
Your environment
- Routed Results Yours
- Constraints & Rules Yours
- Keys & Access - SSO / SAML Supported
DeepPCB
- Engine deployed inside - on-prem / private Available
- Hosted tier: job metadata only Per ToS
Your Tools Stay Your Tools.
Find the tools your teams already run, and the path each one takes into the engine: native, through a plugin, or by a third-party format. No one relearns a CAD tool, and no board waits on a migration.
| Design tool | Native | Plugin | 3rd-party format | Status |
|---|---|---|---|---|
| Altium Designer | Yes | No | Yes | Full with minor exceptions |
| KiCad | Yes | Yes | Yes | Full support |
| Specctra DSN | Yes | No | No | Full support |
| Zuken | Yes | No | Yes | Full support |
| EasyEDA | Yes | No | No | Full support |
| Siemens Xpedition | No | No | Yes | Partial support via DSN |
Nothing about your compliance changes.
Component selection, approved-parts lists, radiation and vibration and thermal, regimes like ITAR and RoHS: they live where they live today, with your engineers and your procurement and your ERP. DeepPCB takes schematics, netlist, and board outline to place & route your board. Your program managers approve what they approved before, your engineers sign what they signed before, and nothing about the way you qualify a board answers to us. Where the board file itself is the controlled article, run on-prem: it never leaves your walls.
Depend on It for the Life of Your Program.
A board you start this year will be in production and revision long after it ships, and the tool that routed it has to still be there. DeepPCB is built by InstaDeep, an enterprise AI company that was acquired in 2023 for $680M.
-
Deutsche Bahn
Reinforcement learning to schedule operations at scale, on Germany’s rail network. A similar set of methods route your boards.
-
BioNTech
Co-developed BioNTech’s COVID-19 Early Warning System.
-
NeurIPS · ICML · ICLR
InstaDeep’s research has been peer-reviewed at top conferences since 2018.
Validate It on Your Boards, on Your Rules.
Public benchmarks prove the engine. Your boards prove the fit.
The criteria are signed before the work begins: how much routes, how the time compares to the hours the boards originally took, and whether your engineers accept the results by their own standard. Kickoff, regular check-ins, a mid-pilot review, and a written evaluation at the end. Getting started is days, not a rollout: the lift is the board file.
Scoped together.Run by us.Judged by you.
Try DeepPCB on your own boards with direct enterprise-grade support for your whole team.
Book My Pilot