Sunghee Yun (He/Him)

Co-founder & CTO @ Erudio Bio, Inc., Chief Business Development Officer @ WeStory.ai, Advisor & Evangelist @ CryptoLab, Inc., KFAS-Salzburg Global Leadership Initiative Fellow @ Salzburg Global Seminar, Adjunct Professor @ Sogang University, Advisory Professor @ DGIST, Global Advisory Board Member @ Innovative Future Brain-Inspired Intelligence System Semiconductor of Sogang University, Network Expert @ Gerson Lehrman Group

Abstract

Industrial AI represents a transformative opportunity for manufacturing organizations, with McKinsey research indicating that AI front-runners can achieve up to 122% cash flow improvement by 2030, while laggards face potential 23% declines. The semiconductor industry serves as an ideal starting point for industrial AI implementation due to its advanced digitalization infrastructure, generating massive volumes of data including equipment sensor data (~1TB/day), metrology image data (~10TB/day), and manufacturing execution data (~10GB/day). This data-rich environment, combined with sophisticated processes and significant business impact potential, creates optimal conditions for developing scalable AI solutions that can expand to other manufacturing sectors.

Computer vision and time-series machine learning form the core technical foundations of manufacturing AI applications. Computer vision techniques address critical challenges in metrology and inspection, including pattern classification, defect detection, and anomaly identification using scanning electron microscope (SEM) and transmission electron microscope (TEM) images. Advanced image restoration methods, segmentation algorithms, and unsupervised anomaly detection enable automatic measurement of critical dimensions with sub-nanometer precision. Meanwhile, time-series ML applications focus on virtual metrology, yield prediction, predictive maintenance, and root cause analysis, utilizing semi-supervised learning and Bayesian inference methods to handle the temporal nature of manufacturing data.

The implementation of industrial AI faces unique challenges that distinguish it from traditional AI applications, including data characteristics such as concept drift, covariate shift, fat data scenarios, and quality issues. Success requires a data-centric approach emphasizing domain knowledge integration, close collaboration with manufacturing experts, and development of fully customized algorithms rather than off-the-shelf solutions. Virtual metrology exemplifies a successful industrial AI application, enabling measurement of unmeasured materials using equipment sensor signals and providing credibility intervals for predictions, effectively equivalent to investing in 100x measurement equipment capacity while optimizing resource allocation for maximum manufacturing impact.

inAI

The rapid advancement of deep learning (DL) technology has catalyzed the proliferation of AI applications across diverse domains such as computer vision (CV), natural language processing (NLP), recommendation systems (RecSys), and reinforcement learning (RL). These advancements have profoundly influenced various facets of daily life, fueling competitions in self-driving car development, facilitating customer service through chatbots, generating creative content, and enhancing e-commerce experiences through personalized recommendations.

Despite these strides, the successful integration of machine learning (ML) techniques into industrial sectors has been limited, hindered by numerous challenges. Manufacturing environments, for instance, are characterized by data drift and stringent accuracy requirements, posing obstacles for conventional ML methods. Additionally, industrial applications often lack accurate labels or rely on subjective human judgments, further complicating the deployment of AI solutions.

This seminar introduces industrial AI as a burgeoning field with vast potential to deliver tangible business benefits and streamline engineering processes. Through case studies, we illustrate successful ML implementations in challenging industrial contexts, highlighting strategies for overcoming barriers. We delve into time-series ML algorithms and their pivotal role in manufacturing, emphasizing their relevance in handling the ubiquitous time-series data generated on production lines.

Furthermore, we address the dichotomy between domain-specific applicability and the genericity and reusability of ML algorithms and software components. Balancing these factors is essential for both our research endeavors and the timely delivery of value to customers. We underscore the importance of aligning customer needs, product roadmaps, software systems, and ML algorithm development for the effective development and deployment of industrial AI solutions.

In conclusion, this seminar emphasizes the imperative of aligning customer-centric values, product strategies, and technical development efforts to drive the successful implementation of industrial AI solutions.

genAI

The contemporary landscape of artificial intelligence (AI) has undergone a profound transformation, driven by the emergence and widespread application of large language models (LLMs), with OpenAI’s ChatGPT based on GPT-x models serving as a pivotal catalyst. This seminar, tailored for university faculty and students, delves into the dynamic intersection of LLMs and generative AI (genAI), engaging both major industry players and agile startups in a comprehensive exploration.

The first part of the seminar initiates with a concise overview of LLMs, followed by a nuanced technical analysis focusing on the attention mechanism, a linchpin within the Transformer architecture. This examination aims to unveil the intricacies underlying the training and inference processes of LLMs having Transformer components, offering valuable insights into their remarkable capabilities.

Expanding the discourse to genAI, encompassing both technical and business dimensions, the seminar illustrates how LLMs and genAI collectively herald a new era. Numerous examples showcase how these technologies stimulate the creation of innovative applications and transformative business models. The first part of the seminar culminates with an exploration of current trends in genAI, providing strategic insights to navigate market dynamics and foster the development of impactful products.

The second half of the seminar shifts focus to critical non-technical topics surrounding the rapid progress of AI development. Engaging in meaningful discussions, we explore the rationale behind pursuing human-level AI, potential biases, ethical and legal challenges, and the elusive concept of AI consciousness. The speaker encourages active participation from the audience, fostering debates on these intriguing topics. The seminar concludes with the speaker’s contemplation on the knowledge, reasoning, and belief systems of AI, particularly those inherent to LLMs, offering a thought-provoking finale to this exploration of the frontiers of AI.