Get Started
Menu

Your Data, Your Deep Learning

Deep learning is emerging as an important success factor for the semiconductor manufacturing business. In the deep learning era, protecting your data is critical, so every company needs to do deep learning on its own data. D2S TrueMask® DLK is a quick-start kit that accelerates the ramp-up process on deep learning for any company involved in semiconductor manufacturing, including fabs, mask shops, equipment or service suppliers, and fabless design companies.

TrueMask® DLK Jumpstarts Your Deep Learning

Deep learning has broad applicability in semiconductor design and manufacturing, but it requires significant training with a large amount of data covering a broad range of geometric configurations. Training with SEM pictures alone is impractical, so augmenting training data through simulation is the only way to make deep learning robust. In addition, all real mask or wafer data is curvilinear. Simulation-based generation of training data therefore requires curvilinear processing. GPU acceleration enables fast full-chip processing of curvilinear simulation of mask and wafer data. TrueMask® DLK from D2S combines all of these capabilities to enable deep learning for photomask and semiconductor manufacturing.

TrueMask® DLK includes:

  • D2S TrueMask® MDP GPU-accelerated, model-based mask data preparation (MB-MDP) solution for complex masks, which include native curvilinear mask and lithography simulators, as well as geometry libraries and verification checks
  • Additional mask and wafer simulation capabilities from D2S partners, including a rigorous eBeam simulator from aBeam Technologies, and a rigorous ArF/EUV lithography simulator that includes mask 3D effects from Fastlitho
  • Pre-trained deep learning neural network model for mask and wafer fabrication
    D2S Computational Design Platform, now the sixth-generation CDP using the NVIDIA Tesla V100 GPU accelerator

Benefits

  • Maintain control of your data by doing your own deep learning
  • Accelerate your deep learning efforts with a complete deep learning kit, including hardware and software solutions
  • Create robust deep learning applications quickly with neural networks pre-trained for semiconductor manufacturing applications

TrueMask® DLK

Deep Learning Kit for Semiconductor Manufacturing

“Our GPU-accelerated lithography simulator called Quantum™, OEMed through D2S, is a rigorous full-feature simulator that works 1000x times faster with better accuracy than any other tool in the industry. It works for EUV as well as for 193i. By having this capability, customers of TrueMask® DLK will be able to train their deep learning networks with accurate simulated data. With simulated data, any type of training data, including many anomalous situations, can be generated accurately and quickly for training. For example, D2S TrueModel® process modeling capability can be used to generate accurate predictions of EUV mask shapes, which are then simulated for EUV lithography with Fastlitho’s Quantum simulator, together with D2S resist and etching model, to generate accurate predictions of pattern shapes on the wafer. After that, SEM or eBeam inspection images can be simulated with aBeam Technologies’ aSEM or Chariot simulator.”

Michael Yeung

President, Fastlitho

SPIE Photomask Technology 2019 Paper: Making Digital Twins Using the Deep Learning Kit (DLK)

Linyong Pang, Suhas Pillai, Thang Nguyen, Mike Meyer, Ajay Baranwal, Henry Yu, Mariusz Niewczas, Ryan Pearman, Abhishek Shendre, Aki Fujimura, “Making digital twins using the Deep Learning Kit (DLK),” Proc. SPIE 11148, Photomask Technology 2019 (October 21, 2019). DOI: https://doi.org/10.1117/12.2538508