“Artificial Intelligence” (AI) has received a lot of attention in recent years and is having a transformative impact on several industries, including the semiconductor industry. The emergence of AI and other related technologies such as machine learning (ML) and deep learning (DL) has brought new perspectives to the semiconductor industry, leading to increased productivity, efficiency, performance and reduced costs. rice field. This article will focus on the different ways AI implementations are making major advances in semiconductor and integrated circuit technology.
Image credit: Gorodenkoff/Shutterstock.com
Implementing machine learning (ML) in the semiconductor industry
The introduction of ML in the semiconductor industry is highly beneficial, especially in terms of predictive maintenance, defect detection, and process control. An article published in the journal Chip focuses on the role of ML in data analytics in the semiconductor industry.
Researchers in semiconductor materials analysis and manufacturing are working hard to develop and implement a variety of new ML algorithms to make better use of vast amounts of data. According to the article, the majority of machine learning (ML) tasks related to semiconductors mainly involve prediction of classification, various material properties, device performance, and manufacturing processes. As a result, most of his ML applications in this field are primarily based on supervised learning.
ML implementations in semiconductor research can be divided into two distinct categories. One is the macroscopic information of materials (lattice constant, space group, elastic modulus, etc.) and the development of data models for all material locations and types. atom. These studies make it possible to predict macroscopic and microscopic properties based on information about each atom, such as machine learning interatomic potentials (MLIPs) and MLFFs (machine learning force fields).
To identify and classify microchip defects, researchers have proposed an automated defect classification (ADC) method. The primary input is an image captured by a scanning electron microscope. Using a convolutional neural network (CNN) approach, this system can detect virtually without human assistance, as experiments demonstrate the high effectiveness of this method.
The limits of ML in the semiconductor industry
Machine learning is a type of computational artificial intelligence. Some flaws are mainly due to inadequate statistical analysis. In materials science and semiconductor manufacturing, there is a lot of data, but it may not be enough to build general ML models, or it may not be effectively leveraged for a particular study. As the generality of the model increases, the amount of training data required to obtain an effective model increases. The degree of abstraction in your ML goal should scale with the amount of training data.
Some results in this area lack statistical analysis of results. When applying machine learning to semiconductors, it is difficult to reliably predict his IID conditions between training data and scenarios.
Insufficient testing in real simulations and production processes is also a major limitation. Much work remains to be done, as much of the research in this area is still in the conceptualization and validation stages.
Scalable Yield Prediction Powered by AI
Manufacturing semiconductor devices involves hundreds of operations, and many factors affect device yield. A recent article in Applied Sciences introduces explainable artificial intelligence (XAI) that takes a scalable input database approach to incorporate different factors into predictions and uses interpretative models to modify manufacturing conditions. I’m targeting this issue by implementing:
Ten different machine learning algorithms are optimized and compared to select the data prediction model with the best performance. The selected RF model shows an enhanced prediction score of 0.520 for mean absolute error and 0.648 for root mean square error. The results of this study have important implications for improving production yields of multiple products using diverse manufacturing data.
Resolving supply chain bottlenecks and chip quality with AI
A new article in the International Journal of Research Publications and Reviews focuses on implementing AI to solve several problems related to semiconductor manufacturing.
It makes sense to build machine learning models using real-time data extracted from existing datasets in electronic systems. This enables companies to gather knowledge from real data rather than relying on assumptions in the planning process, improving forecasting accuracy, developing ahead of customer demand, and using production capacity more effectively. By using it, you will be able to respond quickly. Machine bottleneck problem.
In addition, AI can be used to address product quality issues caused by voids that occur in current inspection and actual visual inspection procedures and capabilities. Current approaches rely on manual labor, such as adding manual inspection gates, which not only lengthens the manufacturing cycle, but also increases manufacturing costs.
Image credit: jiang jie feng/Shutterstock.com
The proposed AI-based solution uses historical data sources for training artificial neural network ML models along with image processing algorithms, and is flexible for more accurate product inspection using dynamic reference photos for comparison. The idea is to create a decision-support tool. In the proposed AI solution, a series of image preprocessing techniques are applied to images from automated optical inspection technology sent to an FTP server, and enhanced and inpainted using color segmentation to extract features. Even the slightest imperfections are detected. Defects on semiconductor chips can be easily detected.
Genetic Algorithms for Root Cause Disorders
The journal Scientific Reports has published an article aiming to develop models that can predict the outcome of a disability based on the salient features of the disability description. A genetic algorithm (GA) is a subset of AI that simulates the method of natural selection. GA performs search operations in complex, large-scale, multimodal environments and provides near-optimal solutions. GA-DT refers to the combination of article genetic algorithm and decision tree classifier. First, the failure analysis description X is preprocessed. Phase 2 presents a word vectorization technique Word2Vec. For vectorized preprocessed data, Phase 3 demonstrates the use of GA variable selection techniques combined with decision tree or support vector machine supervised learning.
The algorithm displays the three most accurate predictions of the fault assessment conclusion for each text sample. The BLEU score quantifies the similarity between the predicted data and the original collection. The salient features selected by the proposed GA-DT method provide the most accurate predictive model of failure conclusion based on failure process descriptions.
market analysis
According to a report published by McKinsey and Company, AI/ML has an opportunity to create enormous commercial value for semiconductor companies. AI/ML is currently bringing in $5 billion to he $8 billion in revenue annually. This is notable, but only about 10% of the maximum industry value for AI/ML. AI/ML could generate $35-$40 billion within the next 2-3 years. This indicates that the semiconductor industry has reached an inflection point, and companies that do not commit significant resources to their AI/ML strategy risk falling behind.
AZoM Deep Dive: Can Blockchain Technology Better Plan the Semiconductor Supply Chain?
References and further information
Ramal, A. other. (2023). Root Cause Prediction of Failures in the Semiconductor Industry, Genetic Algorithms and Machine Learning Approaches. Sci Rep 13, 4934. Available here: https://doi.org/10.1038/s41598-023-30769-8
McKinsey & Company, (2021). Augmenting AI in AI-enabling Fields: Lessons for Semiconductor Device Makers. [Online]
Available at: https://www.mckinsey.com/industries/semiconductors/our-insights/scaling-ai-in-the-sector-that-enables-it-lessons-for-semiconductor-device-makers
Lee Y., Roh Y. (2023) Scalable Yield Prediction Framework Using Explainable Artificial Intelligence for Semiconductor Manufacturing. applied science. 13(4):2660. Available from: https://doi.org/10.3390/app13042660
Liu et al. (2022). Machine learning for semiconductors. chips. 1(4): 100033. Available from: https://doi.org/10.1016/j.chip.2022.100033
Lhasa et al. (2023). A review of semiconductor manufacturing using AI technology. A review of semiconductor manufacturing using AI technology. 4(3): 1376-1330. Available from: https://ijrpr.com/uploads/V4ISSUE3/IJRPR10490.pd
