GAN-SRAF: Subresolution Assist Feature Generation Using Generative Adversarial Networks

Re-examining VLSI manufacturing and yield through the lens of deep learning

“…Although this approach has demonstrated significant speedup compared to model-based approaches while achieving comparable results in terms of process variation band, there is still significant room for improvement by leveraging recent advancement in the field of image processing in computer vision [15], [31]. Recently, we proposed two GAN schemes to address the SRAF generation task [19]. In the first, we propose to use CGAN for SRAF generation by casting the problem into an image translation task where the two images domains are: (i) original layout and (ii) layout with SRAFs.…”

“…In addition, extracting the SRAF information from the image to be mapped back to the layout file can be prohibitively expensive. Hence, a special encoding scheme, typically used in keypoint estimation [58], [59], is proposed in [35] to overcome the aforementioned limitations. The proposed scheme is based on multi-channel heatmaps which associates each object type with one channel in the image [59], [60].…”

“…Here also, DL can be used to advance RETs by presenting its successful application in sub-resolution assist feature (SRAF) generation which is a key RET adopted to improve the target pattern quality and lithographic process window. In our recent work [19], a novel formulation for SRAF generation was proposed to cast the problem as a domain transfer task. Then, GAN-SRAF was proposed for efficient SRAF insertion leveraging recent advancements in generative adversarial learning.…”

Full-chip thermal map estimation for commercial multi-core CPUs with generative adversarial learning

“…Recently GAN-based methods have been applied for VLSI physical designs such as generation of the various noise maps to facility the IR-drop noise sensor placement [38], for layout lithography analysis [39] and sub-resolution assist feature generation [40], for analog layout well generation [41]. But less studies have been investigated for data-driven circuit level and thermal analysis to model the dynamic systems described by the partial differential equations.…”

Powernet: SOI Lateral Power Device Breakdown Prediction With Deep Neural Networks

“…Recently, machine learning techniques have been widely adopted for different applications in the field of electronic design automation [13]- [18]. It has been widely used for its capability in learning hidden correlations between the electronic parameters and the electric performance with high accuracy and efficiency [19]- [21].…”