I’m very happy to have another guest post to publish ! During the last SEMI Fab Owners Alliance (FOA) meeting in Portland, ME
David Meyer, co-founder and CEO at Lynceus AI and
Ariel Meyuhas, COO and Founding Partner at MAX Group
presented a very interesting approach using AI to improve equipment maintenance. I think this is a really nice approach to tackle a long existing problem in the industry. I approached David and Ariel to share there work here – Enjoy reading !
AI-enabled Precision Maintenance
This blogpost follows a presentation made at SEMI FOA session in October 2024. The slides are attached below.
AI-Enabled Precision Maintenance: a new way of managing capital capital equipment
As you all know, servicing tools becomes increasingly complex and PM checklists keep growing. If we keep thinking about maintenance in the same way, we risk degrading COO and profitability. On the brighter side, there is now abundant data to describe equipment behaviour and the technologies that can leverage this data are getting more and more mature.
This is an opportunity to provide a step-change in equipment productivity.
Current Maintenance Paradigm: long, rigid and blind
Here is the problem: we are used to fixed PM schedules. At every PM, we run through the same list of actions – irrespective of what’s actually happening to the tool / process.
As a consequence, our PMs are long, rigid and blind.
This negatively impacts PM downtime, qual complexity, spare parts consumption and even unplanned downtime. The time we spend on unnecessary interventions is time we could have spent troubleshooting more accurately or implementing longer term fixes.
Now, what if we could run PMs differently? what if we could do only what is necessary, when it is necessary?
This is what AI-enabled Precision Maintenance aims for.
A new maintenance concept
We came up with the concept of a dynamic and fractional PM checklist based on real-time status of the tool and of the process. Concretely, for each part of the tool and at any given time, we can recommend whether or not an intervention is required.
We formulate this recommendation based the tool’s maintenance history and on its most recent behaviour.
With AI-enabled precision maintenance, you’d be able to access real-time recommendations, for each part of the scope: how many wafers can I still produce before change? is the tool behaving normally or not?
This tool has the power to optimize PM scheduling and inform the engineer on which parts of the PM scope to execute or not.
This translates into:
- Shorter G2G
- Simpler PM, and therefore simpler quals
- Reduced spare parts consumption
- Improved team productivity
As a fab manager, it gives you the possibility to optimize for COO, productivity and capacity simultaneously. The concept was tested on a bottleneck toolset with manual extraction of data, which is very labour intensive.
Combining expertise
Rolling-out AI-enabled precision Maintenance successfully requires a combination of different skills & expertise working together.
First, deep Maintenance & Equipment expertise: MAX Ops defined the Precision Maintenance concept and demonstrated its impact through manual implementation. We know it works, but it is still more empirical than systematic.
Then, we need to make it scalable. AI techniques help extract and leverage information from equipment data automatically, which then feed dynamic PM checklists.
The last piece of this puzzle is of course the users in the fab. Adopting such a novel approach to capital equipment management requires a true partnership with the fab, in order to define the best way to validate, deploy and use this tool.
Impact on Operations
While AI-enabled precision maintenance is a new concept, Precision Maintenance is not.
At MAX Engineering, we have been supporting fabs in optimizing their maintenance cycles by defining more frequent, more fractional scopes – with some great results.
In this example, we were working on a set of Litho tools in a 300mm fab.
Precision maintenance helped this fab reduce G2G time by 25% on average and improve M-Ratio from 3 to 5:1. On an annualized basis, this means 4% net uptime gains.
Now this is what you can obtain by breaking down PMs into smaller fixed scopes, through a highly manual and empirical process.
So what is the expected impact of boosting Precision Maintenance with AI?
First, we would be able to define dynamic PM scopes, updated in real-time based on the tool’s most recent behaviour. We expect this can double the productivity gains evidenced with Precision Maintenance, parts & people combined.
Second, we could replicate this approach seamlessly to other bottleneck areas in the fab, 6 times faster than if we had to redo it manually. At scale, this means turning local uptime gains into realized throughput expansion.
Looking forward: AI-Module Engineer Assistant
To conclude, I’d like to give you an idea of what the future of module engineering can look like.
We saw that AI-Enabled Precision Maintenance will be a step change in maintenance productivity, but this is only step 1.
We are already working on an AI module engineer assistant, able to perform most of the routine tasks around process & equipment management. We imagine a tool that can:
- Flag drifts in equipment behaviour
- Define maintenance checklists based on the tool’s most recent behaviour
- Publish and share reports on recent PMs
- Build correlation studies around events of interest
- Support technicians during interventions
This is step 2, and we’re working on it today.
Here is the complete presentation from the SEMI FOA meeting:
If you are interested to learn more, please contact David or Ariel directly:
David Meyer : david.meyer@lynceus.ai
Ariel Meyuhas : ariel_meyuhas@maxieg.com