Multi-beam mask writing was beginning to move from forecast to reported production activity in the semiconductor industry, according to the eBeam Initiative’s 2019 surveys. The findings, drawn from responses across the semiconductor ecosystem and data supplied by photomask manufacturers, also highlighted challenges around advanced-node mask turnaround time (TAT), EUV mask yield and the expected role of deep learning in purchasing decisions. These results offer a time-specific view of how the industry assessed manufacturing priorities in 2019—not a statement of current market conditions.
Multi-beam mask writing enters the production picture
The mask makers’ survey covered activity from July 2018 through June 2019 and collected responses from 11 captive and merchant photomask manufacturers. Together, participants reported 599,536 masks. The survey indicated that 30 percent of reported masks were written using electron-beam variable shaped beam (VSB) technology. It also documented masks being written with multi-beam technology, marking the first time the survey identified multi-beam as a pattern generator.
The distinction matters because mask writing is part of the pattern-generation process used to create photomasks for semiconductor manufacturing. The survey did not present multi-beam as having displaced VSB; rather, it showed that both approaches were represented in reported activity. Looking ahead, respondents predicted that multi-beam systems would account for 50 percent of all new eBeam mask writers purchased by the end of 2022. That figure was a forecast made in the 2019 survey, not a measurement of purchases that later occurred.
Advanced ground rules increase turnaround pressure
Mask turnaround time was another notable finding. The reported average TAT for masks at 7-nm to 11-nm ground rules was 11.07 hours—more than twice the average for masks at 32-nm to 45-nm ground rules. The report also noted that TAT continued to be significantly higher for advanced ground rules.
For production and engineering teams, the figures underline a practical planning concern: mask timing can vary substantially with the ground rules involved. The survey also reported an average write time of 8.64 hours for VSB writers. Write time and overall mask TAT are related operational measures, but they are not interchangeable; the survey presented them separately. In capacity planning, teams should therefore distinguish the time spent writing a mask from the broader turnaround measure used in the survey.
Mask process correction (MPC) was reported for 44 percent of masks at ground rules below 11 nm. MPC addresses mask-related effects in pattern formation, and its reported use at these ground rules points to its role in advanced mask workflows. The survey also reported EUV mask yield at 74 percent and counted 2,789 EUV masks during the covered 12-month period.
EUV readiness and lithography expectations
Perceptions survey responses showed positive expectations for several elements of EUV adoption. Eighty-two percent of participants predicted that EUV pellicles would be available for high-volume manufacturing (HVM) by 2022. A pellicle is a membrane used with a photomask; the survey focused on whether EUV pellicles would be available for high-volume production, rather than reporting that this milestone had already been achieved.
Views on actinic mask inspection were also broadly positive. Only 5 percent of respondents predicted that actinic inspection would never be used in HVM. The survey separately found that 75 percent expected inverse lithography technology (ILT) to be used alongside EUV lithography on at least one layer of a high-volume production chip by 2022.
Respondents also perceived that ILT use in production had increased compared with the previous year’s survey. The share indicating that no production layers were yet using ILT fell from 22 percent to 12 percent. These responses describe industry perceptions at the time and should not be read as a direct measure of production adoption.
Deep learning enters purchasing discussions
The perceptions survey added questions about deep learning and semiconductor operations. Sixty-one percent of respondents said it was somewhat likely to very likely that their organizations would purchase products or services based on deep-learning content by the end of 2020. When asked about their customers’ purchasing decisions, 76 percent gave the same likelihood range.
The results suggest that, in 2019, industry participants expected deep learning to become relevant to purchasing criteria across semiconductor operations. The survey did not identify specific applications, products or performance outcomes, so the findings are best understood as a signal of anticipated buying interest rather than evidence of a particular deployment.
What the survey means for semiconductor teams
Taken together, the findings describe an industry balancing new pattern-generation approaches with demanding mask requirements. Multi-beam writing had entered reported mask activity, while respondents expected its share of new eBeam writer purchases to grow. At the same time, advanced-ground-rule masks showed longer turnaround, and the survey documented ongoing use of MPC and activity in EUV mask production.
For engineers, integrators and operations leaders, the practical takeaway is to assess mask-writing technology, process correction, inspection and EUV readiness as connected planning topics—not isolated equipment choices. The survey’s forecasts can help explain the expectations circulating in 2019, while its measured results provide a snapshot of reported activity for the period it covered. Any present-day decisions should rely on current operational data rather than treating those historical expectations as outcomes.