Multivariate timeseries anomaly detection for advanced manufacturing process- ON-425Desired discipline(s): Engineering - chemical / biological, Engineering, Engineering - other, Computer science, Mathematical Sciences, Mathematics, Statistics / Actuarial sciences, Chemistry, Natural Sciences, Physics / Astronomy
Company: Solid State AI
Project Length: 4 to 6 months
Preferred start date: 04/01/2021
Language requirement: English
Location(s): Toronto, ON, Canada
No. of positions: 1
About the company:
Solid State AI works with advanced manufacturers to optimize yields and throughput by building machine learning enabled software.
Please describe the project.:
We are looking to push the frontiers of anomaly detection technology for a state-of-the-art multi-step semiconductor fabrication process. We are developing highly robust and autonomous algorithms which, given a small number of examples of nominal tool data, can predict anomalies with high precision. The challenge is that the input time series data is multi dimensional, and so our approach must capture the relationships that exist between variables. The goal is to effectively learn the nominal range of process parameters and tool operation from a multi-modal distribution of process parameters. Thus, the solution must be able to learn the intended sequence of processing steps in the process directly from the data.
- Strong machine learning and software development skills is a must.
- A background in physics or a relevant engineering domain is helpful, but not a requirement.
- Strong independent research capabilities are required, as there will be substantial learning on the job.
- In addition, our ideal candidate is able to effectively experiment with novel machine learning approaches and converge towards promising solutions under significant time constraints.
- The candidate must also be able to clearly articulate results to colleagues.
- We will also be working with the candidate to produce a reusable code base and so experience with version control, testing and deploying models are important.