This year marks the 40th anniversary of the establishment of Fanlin Group. In the past 40 years, we have insisted on continuous innovation and breakthrough, and have made pioneering progress, including the exploration of equipment intelligence. Looking ahead, we believe that for the semiconductor industry, device intelligence will be the guiding light for future developments.
Increasing difficulty and cost of manufacture
40 years ago, it was hard to imagine that semiconductor technology would bring about such a dramatic change in the world. In 1980, for example, the microprocessors in early personal computers contained about 50,000 transistors, while today’s advanced chips typically contain more than 30 billion transistors.
However, the rapid development of technology also brings great challenges.new generationdeviceThe development of the technology means more complex structures, more process steps, new materials and design rules, and the dependence between parameters is also increasing.In this context, our development time and costs increase time and time again to bring innovative products into mass production and ultimately into the hands of consumers. Today, the semiconductor industry has become one of the industries with the highest R&D intensity. At the same time, we are also responsible for the environment, employee well-being and community health, and must advance semiconductor technology in a sustainable manner to change the world.
Device intelligence is key to meeting the challenge
Device intelligence can enable a virtuous cycle of data and learning processes to help us address these challenges. By leveraging its powerful data collection and processing capabilities, we can achieve innovation and breakthroughs in chip manufacturing.
Lam’s vision for equipment intelligence is to bring data modeling, virtualization and artificial intelligence into every link, including design, development, procurement, construction and support systems and processes, from conceptual design and feasibility study to realization mass production.The ultimate goal of device intelligence is to reduce costs, resource consumption and waste while accelerating the speed of technological transformation, that is, to unleash innovation by removing barriers of complexity.
In the era of intelligent equipment, the digitization of the whole process is the key. Every step of any system from initial concept to termination of operation should leave a footprint of data.
It has become very difficult to achieve revolutionary innovations in chip design and manufacturing processes. In order to meet performance and cost requirements, we often need to search a needle in a haystack among trillions of potential formula combinations to find the best formula.
Take etching for example. The cost of process recipe development and validation has grown tenfold over the past decade,In this context, virtual process development is an important problem-solving tool that helps engineers find the best recipe faster.
Despite our growing understanding of specific processes, purely physical models of etching and deposition processes are too complex and computationally intensive to produce practical results.Our strategy is to combine machine learning and historical data mining to build a computable model that operates on the model to find the right combination of variables and successfully find the best solution among 10^14 choices.By working closely with chip manufacturers on structure and process requirements, we have leveraged virtual process development to optimize primary combinations and successfully reduce the number of experiments required and cost by more than 20%. The method is still in the early experimental stage, and we will need to continue to work closely with chip manufacturers to scale up.
The challenges posed by complexity are not limited to R&D. To achieve mass production of chips, manufacturers must replicate the optimal process in hundreds of chambers, guaranteeing the production of hundreds of thousands of wafers per month to angstrom tolerances. To ensure consistent performance across all processes, we need to invest in smarter equipment and services.
This requires intelligent devices with self-awareness, self-adaptation and self-maintenance capabilities.
·Self-awareness means that a device understands its own software and hardware configuration and can monitor key performance indicators using sensors. Compared to the previous equipment, the new generation of equipment adds about 400 sensors, and the data they generate, combined with information from the manufacturer’s fab, can reduce the time for chamber matching from weeks to days.
·Adaptive capabilities mean that these smart devices can continuously adjust themselves or optimize productivity based on unit process deviations and material changes.
·Self-maintenance functions, including automatic calibration, cleaning or maintenance, help increase machine output. The self-maintenance capability of a group of equipment running a critical application in a fab has been proven to save customers 45,000 man-hours per year.
Data-based intelligent service ecosystem
Intelligent equipment requires an intelligent service ecosystem, otherwise the overall equipment efficiency of hundreds of process chambers cannot be guaranteed, and such intelligent services are inseparable from data.
To find corrective actions, predict maintenance events, manage smart parts, and optimize equipment performance, we need to develop algorithms specifically based on data.To this end, Lam Group has established a device data repository, and has opened up data control and access rights to customers according to the situation while maintaining data integrity. This flexible solution helps us meet the growing challenge of maintaining these complex systems in a volume production environment, ensuring that equipment yield targets are met while meeting tight process tolerances.
Additionally, virtual and augmented reality technologies can leverage data to help engineers become more efficient and share the expertise of experts around the world. Experts can use relevant tools to provide engineers with remote training and support, real-time supervision and guidance with the help of live video.The COVID-19 outbreak has also given us a first-hand look at how these remote support technologies can be business saviors. With travel significantly reduced, the need for remote support will increase rapidly.
Collaboration is critical to continuous innovation and is the key to realizing all the benefits of device intelligence.Today, the semiconductor industry is in a loosely separated model, and the industry must transition into a network of nodes to achieve the flow of data and the integration of algorithms from all parties, so as to achieve more effective industry control.
We believe that by unleashing the full potential of device intelligence, we can foresee the future of the world.