In a new study, Lam Research Corp., Fremont, Calif., examined the potential for the use of artificial intelligence (AI) in process development for chip fabrication and found that a \u201chuman first, computer last\u201d approach can reach process engineering targets dramatically faster and at half the cost compared to today\u2019s approach.<\/p>\n
The study, recently published in the journal Nature<\/em>, identifies an opportunity to address two grand challenges facing the industry: reducing development costs and accelerating the pace of innovation to meet the increasing demand for next-generation chips.<\/p>\n The rising complexity of next-generation chips continues to drive process development to be more challenging and expensive. Seeking a more efficient approach, researchers at Lam put talented process engineers head-to-head against AI-enabled computer algorithms in the study.<\/p>\n To manufacture every chip or transistor designed, experienced and skilled engineers must first create a specialized recipe that outlines the specific parameters and permutations needed for each process step. Hundreds of steps are required to build these nanometer-sized devices on a silicon wafer. Process steps typically include multiple instances of depositing thin layers of materials onto silicon wafers and etching away excess material with atomic-scale precision.<\/p>\n This essential phase of semiconductor development is currently done by human engineers, largely using their intuition and a \u201ctrial and error\u201d approach. With every recipe unique to the chip design and more than 100 trillion possible options to incorporate, process development can be laborious, time-intensive, and costly \u2013 increasingly slowing down the time needed to achieve the next technology breakthrough.<\/p>\n In the Lam study, machine and human participants competed to create a targeted process development recipe at the lowest cost, weighing a variety of factors associated with test batches, metrology and overhead expenses. The study concluded that while humans excelled in solving challenging and out-of-the-box problems, a hybrid human first, computer last strategy can help address the tedious aspects of process development and, ultimately, speed up process engineering innovation.<\/p>\n \u201cAlthough critical to the creation of each and every chip produced, the plasma physics of process engineering has been for decades rooted in the same scientific approach that Thomas Edison used: trial and error,\u201d said Rick Gottscho, executive vice president and strategic advisor to the CEO \u2013 Innovation Ecosystem at Lam Research and co-author of the study. \u201cOur research showed that while engineering talent remains essential to innovation, process engineering costs can be reduced by 50 percent by integrating AI at the right stage and with the right data. The study provides a prescriptive approach for bringing together the best of human-led engineering and the best of what data science and machines offer to create a combination that performs better than either one alone. If realized, this hybrid approach can lead to significant savings in both dollars and engineering time for the industry.\u201d<\/p>\n Lam is currently incorporating the key learnings from the study into its development operations. The Lam study provides initial guidance on how to successfully integrate human knowledge, skill and experience with AI\u2019s ability to rapidly assess numerous possible combinations in process engineering.<\/p>\n \u201cBy complementing engineering expertise with AI using the human first, computer last approach, the tedious and laborious aspects of design are alleviated for engineers, freeing them up to focus on the creative areas of development and explore innovations that may have been out of reach either due to bandwidth or cost,\u201d said Keren Kanarik, technical managing director of Lam Research, lead author of the research paper and a former process engineer. \u201cWhile the application of AI in process engineering is still in its infancy and human expertise and domain knowledge is essential for the foreseeable future, the results point us to a path to foundationally change the way processes are developed for manufacturing chips.\u201d<\/p>\n