For the past 12 months, Samsung’s foundry division has occupied a leadership position at the 14nm node. The company recently revealed some of its plans for future nodes, including an aggressive rollout of extreme ultraviolet lithography (EUV), a technology that’s been promised, hyped, and slowly developed for nearly twenty years at this point.
The image above shows why companies continue to claw for EUV lithography, despite its problems, power costs, and vastly increased cooling requirements. Current technology is based on 193nm wavelengths of ArF excimer lasers, but this technique can no longer scale to the tiny feature sizes used in cutting-edge lithography. One solution to this problem is to use multipatterning, in which the same wafer runs through additional exposure steps to create the final etch. The problem with multi-patterning is that it’s both difficult and expensive — and it slows the total wafer throughput of the fab. If etching a wafer used to require 60 masks and multi-patterning increases that to 90 masks, there’s a significant throughput hit and a greater chance that defects will be introduced at some point in the manufacturing process.
Extreme ultraviolet lithography uses 13.5nm wavelengths of ultraviolet light to image much finer features and avoid the cumbersome and expensive multi-patterning process. The reason EUV has taken so long to come to market is that working with wavelengths of light this small is incredibly difficult — there have been myriad problems related to defect density, the need to work in near-vacuum conditions, difficulties with the lasers themselves, and the lack of a suitable pellicle, or thin film membrane that protects the photomask from particle contamination. All of these issues are why Intel and TSMC have both implied they won’t introduce EUV until they hit the 5nm node.
Samsung, in contrast, is forecasting EUV adoption at 7nm. If the company hits that target, it would give it a significant potential cost advantage over its foundry rivals — but only if the EUV hardware is ready to come online. Betting and missing on EUV as a 7nm solution could cripple Samsung’s foundry technology if TSMC and Intel bet on traditional scaling methods and see better results.
Samsung is announcing a number of near-term improvements and changes to its 2016 product roadmaps, including:
A few notes on these milestones: When a foundry says its ramping a product solution, that doesn’t mean that technology is actually ready for shipping customer designs yet. There’s a significant lag time depending on product family, foundry space, and customer validation — Apple bought early 14nm production from both TSMC and Samsung, while other smartphones and graphics cards took 6-12 months to transition to the new node. Samsung may well be ramping a second-generation 10nm node, but that doesn’t mean end users will be buying 10nm-class equipment in six months.
Samsung is generally offering more tools, ecosystem options, and capabilities like 2.5D interposers for HBM and reference design flows. These types of features are important to many different types of customers and investments like this point to Samsung taking the foundry market seriously.
As always, keep in mind that the term “node” is used somewhat differently by the different foundries. Intel’s 14nm and 10nm nodes are true feature shrinks across the board, while both Samsung and TSMC used a hybrid approach for their 16/14nm products. Samsung’s 10nm is thought to be a hybrid between the 10 and 14nm nodes, just as 14nm is a hybridized 14nm / 20nm process.
The question these days isn’t so much whether Intel is technically ahead in the foundry game — it is, and will likely continue to be, even with its new, delayed cadence. The question is whether those gains continue to be worth the costs, and when the foundry node shrink economics fail altogether. An EUV rollout at 7nm or 5nm will push things along a little while longer, but what happens after that is still largely unknown.