Skip to main content
Common questions about Atomscale’s platform, technology, and how it fits into your workflow.

​
How Atomscale Works

Atomscale connects instrument and characterization data, extracts quantitative process information, and helps you investigate growths and understand unexpected behavior. You can monitor incoming data and use the assistant to build analysis notebooks with inspectable code, explanations, and results.We call this the Integrated Process Environment (IPE): a system that brings real-time and asynchronous measurements together for materials processing.
Atomscale operates through three stages:
  1. Connect: Managed desktop integrations, manual uploads, screen capture, and the Python SDK bring instrument data into the platform. Data-specific processing pipelines extract structured results and derived metrics.
  2. Analyze: Adapter and timeseries models provide representations for comparison and analysis. In an analysis notebook, you can ask the assistant to investigate a growth, examine the evidence, and refine the analysis. See Analyze.
  3. Act: Use the findings to choose the next check, record a decision, or plan an experiment. Automatic detection must be enabled for the deployment. Predictive models and control integrations require validation for the intended application.
Adapter models are proprietary pipelines customized to specific data types (e.g., RHEED video, ellipsometry signals, tool sensor logs). They transform raw source data in real time into generalized, comprehensive fingerprints of the data stream. This enables frictionless use of the data for large-scale pattern recognition and automation, without requiring you to build custom analysis for each data type.
Traditional process control relies on parametric statistical models, fixed setpoints, and periodic manual adjustments. Atomscale brings raw instrument and characterization data into an investigation alongside tool parameters, so you can examine relationships that setpoints alone may miss.In one RHEED case study, Atomscale detected a surface reconstruction change 40 seconds earlier than manual observation of the same feed. That result describes the study conditions, not a general timing guarantee.
Process-control integration depends on the deployment and requires validation for the intended use. For everyday investigation, use analysis notebooks.
Results depend on your process and data, but demonstrated outcomes include:
  • Earlier anomaly detection: Detecting nucleating surface reconstructions 40 seconds ahead of manual observation
  • High-accuracy predictions: Surrogate models for wafer uniformity achieving >96% accuracy from recipe and sensor data alone
  • In-situ composition estimation: Correlating diffraction features with ex-situ composition measurements to enable real-time composition predictions on future runs
  • Trial success prediction: Predicting growth success or failure in 90% of cases from an initial set of roughly 10 labeled samples
  • Quantitative layer-by-layer comparisons: Automatically differentiating growth conditions and doping compositions from raw ellipsometry data in ALD workflows
These are separate application-specific results. See Case Studies for the available setup and limitations. They are not performance guarantees for another process or analysis.

​
Supported Tools & Data

Atomscale supports the following deposition methods:
  • Molecular beam epitaxy (MBE): most mature capabilities
  • Chemical vapor deposition (CVD / MOCVD)
  • Atomic layer deposition (ALD)
  • Physical vapor deposition (PVD)
  • Sputtering
  • Atomic layer etch: in active development
The platform is designed to be flexible across tool types. If you’re working with a deposition method not listed here, reach out to discuss your setup.
Atomscale integrates data from a wide range of in-situ and ex-situ characterization techniques:
  • Diffraction: RHEED, XRD, XRR
  • Spectroscopy: XPS, Raman, Photoluminescence, Ellipsometry, OES, SIMS
  • Microscopy: SEM, AFM, Optical images
Our adapter model architecture is designed to extend to new data types. If your instrument generates structured or streaming data, we can likely support it.
Our customers work across advanced materials applications including:
  • Silicon photonics — barium titanate on silicon, perovskite/silicon tandems
  • III-V compound semiconductors — GaN, GaAs, InP, SiC, and combinations for photonics, optoelectronics, and quantum devices
  • Next-generation transistors — controlling chemical composition for 2D FET channel materials
  • Quantum cascade lasers — real-time feedback for dynamic process control across alternating layer stacks
  • Advanced magnets and functional materials
If you’re working on thin film deposition of active electronic or photonic materials, Atomscale can likely help — contact us to discuss your specific use case.

​
Integration & Getting Started

Atomscale works alongside your existing instruments and data sources. Bring data in through managed desktop integrations, manual upload, RHEED screen capture, or a custom Python SDK application. Desktop integrations collect locally while the desktop app buffers and uploads using its signed-in session.Use Integrations in the desktop app for local collection and Gateways in the browser for remote management. See Integrations and Connect.Recipe files, tool APIs, and control interfaces such as SECS/GEM may be part of an application-specific integration. Discuss the requirements and validation with Atomscale rather than assuming that ingesting data enables process control.
We use a three-stage approach:
  1. Proof of concept with historical data: We onboard your existing data at no cost to demonstrate value on your specific process and assess whether the models extract useful information.
  2. Integration with live data: We configure data connections for your environment. Detection and control integrations depend on your application and deployment.
  3. Ongoing use: Monitor growths, use notebooks to investigate results, and preserve findings. Where appropriate, separately validate operator assistance or automated interventions for the process.
Our serial model architecture is designed for exactly this. The adapter models handle tool-specific data transformation, while the foundation model provides generalized pattern recognition. This means the platform adapts to your specific processes without requiring massive retraining. Even with small labeled datasets — as few as 10 samples in some cases — we can produce practical models for continuous tunability over your process design space.
Atomscale offers three deployment options:
  • Cloud (Web UX) — Hosted platform accessible through your browser
  • API — Programmatic integration for automation workflows and custom tooling
  • On-premises — Local deployment for environments with strict data requirements
Your process data stays within the agreed deployment boundary. We work with your security and IT teams to meet your organization’s requirements.

​
Pricing & Engagement

Atomscale uses a usage-based model tied to the volume of data ingested and processed. This aligns our pricing with the value you receive — you pay for what you use, and costs scale with your operations rather than requiring a large upfront commitment.The initial proof of concept uses historical data and is provided at no cost.
Atomscale serves several personas:
  • Process engineers and module owners: Primary users who develop, maintain, and diagnose the process of record. They interact with the platform daily during runs.
  • Metrology engineers: Use the platform to bridge in-situ and ex-situ measurements, building stronger connections between characterization and process.
  • Engineering managers and fab directors: Track KPIs including yield improvement, downtime reduction, and product variance. Typically the decision makers and economic buyers.

​
Company Background

Advanced, application-specific materials are critical for the next generation of technological progress — but compelling materials often get stuck at the lab scale. We can propose new materials faster than we can prove they work in production.Atomscale is building purpose-specific AI to bring real-time feedback, visibility, and automation to advanced materials manufacturing. Our product vision progresses through three stages:
  1. Real-time analysis: A platform to find an edge in your data by using 100% of your signal in real time
  2. Virtual characterization: Intelligence layer models that predict the state of the process relative to past runs or in absolute terms
  3. Self-driving process of record: Adaptive, active process control that optimizes for consistency of output rather than consistency of tool state
By enabling dynamic control of complex process physics, we’re creating a new standard for how advanced materials get made, with the repeatability and precision needed to scale electronics, photonics, quantum technologies, and energy storage.
Atomscale was founded by Chris Price and Jason Munro, two materials scientists with a shared goal of turning decade-long synthesis campaigns into tractable computational problems.Chris Price has spent nearly a decade at the intersection of quantum mechanics, AI, and physics-based modeling. His career spans deep technical research and enterprise data products, focused on translating advanced science into commercially viable technologies. He holds a PhD from the University of Pennsylvania.Jason Munro earned his PhD in computational materials science from Penn State and continued as a staff scientist at Lawrence Berkeley National Laboratory. There, he served as a lead developer of the Materials Project, the world’s largest open database of simulated materials data.
The team brings expertise spanning materials science, physics, AI/ML, high-performance computing, scientific software, and life sciences. Personal backgrounds, advisors, and investors span institutions including Duke, Penn, Berkeley, The Materials Project, Notre Dame, MIT, and the Semiconductor Research Corporation.

​
Getting started

If your data is already in Atomscale, follow Analyze to investigate a growth with the assistant and a notebook. Otherwise, start with Connect. For exact controls, use the notebook reference.
You can describe an investigation to the assistant and ask it to build and refine the notebook. You can also inspect and edit the Python cells directly. Review the selected data, assumptions, and evidence before relying on a conclusion. Assistant features are unavailable to demo accounts.
Identify the growth or data item and explain your observation or question. Include useful context such as an expected outcome, an approximate time, or a known process change. Ask the assistant to explain the evidence and uncertainty. You do not need to select a metric or analysis method in advance.

​
Connecting and finding data

Yes. Use Gateways to inspect workstations and manage supported integration actions remotely. Local Integrations, activity logs, upload queues, and workstation-only prompts are accessed in the desktop app. See Manage Gateways Remotely.
Open Documentation in the desktop app. The bundled guide works offline and covers camera setup, tool logs, connection sharing, and troubleshooting. It is separate from this public website.
Look in Data Catalog, including items without a physical sample using the footer control. Check the item’s processing status. Gateway tool logs do not always have a matching growth sample. See Finding Collected Data.

​
Notebooks and shared work

No. The SDK is preinstalled and the notebook client is authenticated automatically. Use Client() without putting an API key in the notebook. Your own Python application outside Atomscale needs its own SDK setup.
A notebook with saved results normally opens in the result view. Click Edit notebook when available. If another user is editing, you can review saved results while you wait. See Saving, Versions, and Access.
Click Save Changes in the notebook header. New notebooks are visible within their organization, and saved results can be reviewed without taking the live editor. Put the notebook URL and confirmed findings in the relevant sample or project notes. See Saving, Versions, and Access for permissions and version behavior.
Yes. Use Save as Template to create an organization template, then create a separate notebook from it. Review the new inputs and assumptions before using the result. See Reusable Templates.

​
Monitoring and results

No. A notebook investigation is separate from continuous detection and control integration. Automatic detection must be enabled for your deployment. Contact support about availability and integration requirements. See Act and Changepoint Detection.