At the forefront of semiconductor manufacturing, exposure lithography technology has always been a key bottleneck in determining chip performance and process node evolution. ASML, a major Dutch semiconductor equipment manufacturer, has established an almost unassailable market position with its deep ultraviolet (DUV) and extreme ultraviolet (EUV) laser machines. However, the Japanese optical giant Canon is trying to break through with another path, which is Nanoimprint Lithography (NIL). This new pattern transfer technology, considered "non-optical", is being positioned by Canon as a potential disruptor of the next generation of chip manufacturing processes.
The concept of nanoimprint technology can be traced back to 1996. It was first proposed by the University of Texas in the United States and later spawned the commercialization company Molecular Imprints Inc. (MII) in 2001. Canon acquired MII in 2014, officially incorporated NIL into its semiconductor equipment layout, and launched its own technology brand J-FIL (Jet and Flash Imprint Lithography).
Unlike traditional exposure, which uses optical projection imaging, J-FIL uses a "jet, flash and imprint" process. The process starts by spraying and distributing photoresist on the wafer surface, replacing the spin coating mechanism. Then, the photomask (template) with the fine circuit pattern is directly pressed into the liquid photoresist, and then cured with ultraviolet flash to form the pattern. The entire process takes less than a second to complete and can operate at room temperature, significantly reducing energy consumption.
One of J-FIL’s biggest engineering highlights is its innovative “i-MAT (Interference Moire Alignment Technology)”. The system can instantly detect the alignment deviation between the wafer and the reticle through interference fringes, and uses 16 piezoelectric actuators for fine-tuning. It also corrects high-order errors through laser heating to improve overlay accuracy.
Just after Canon promoted NIL, its biggest selling point was its significant advantages in cost and energy consumption. First, the resolution of NIL is theoretically comparable to or even better than that of EUV, and its limit mainly depends on the resolution capabilities of the electron beam writing equipment. Second, Canon estimates that a four-module J-FIL device would cost only one-tenth of the cost of an ASML EUV system. Even taking into account the difference in input and output (NIL is about 100 wafers per hour, EUV is about 220 wafers), its single-wafer process cost is still about a quarter of EUV.
As for energy efficiency, NIL also leads by a large margin. EUV relies on a CO₂ laser system that consumes huge amounts of energy, with a total power of up to 1 megawatt, while NIL equipment only requires about 100 kilowatts, reducing energy consumption by more than 90%. For high-energy-consuming wafer fabs, if NIL can overcome the bottleneck of mass production, it will bring double benefits in terms of manufacturing costs and carbon emissions.
Despite the huge potential for development and application, the industry generally believes that there is still a significant distance between NIL and actual mass production. The key lies in two difficult problems, namely mask life and defect control. Since NIL directly imprints the wafer, the micron-scale and even nano-scale structures on the template are extremely fragile. Current mass production tests show that the life of the photomask can only support about 50 wafers, which is far less than the 100,000-level life of the optical mask. Canon claims that the new design can be extended ten times, but the industry's actual test results are still not ideal.
What’s more serious is that any tiny defects on the photomask will be copied to all wafers, causing a serious “repeated defect” problem. Due to the short lifespan, high cost, and time-consuming testing of NIL templates, it is almost impossible to inspect them one by one. To equip a complete NIL production line with enough inspection machines, the required production capacity is equivalent to the global supply of mask inspection equipment for a whole year, and the economic benefits are obviously inconsistent.
In addition, the NIL template production process itself is also extremely complex. Unlike optical exposure, which can be magnified 4 times, NIL needs to be written at 1 times the size of "main template → sub template → working template", and defects may occur at every step. For feature sizes below 20 nanometers, you need to rely on the support of the most advanced multi-beam electron beam writer (MBMW), which has significant cost and yield pressures.
In addition, in addition to template issues, NIL’s overlay accuracy and productivity still lag behind ASML’s EUV system. Since Canon uses a single wafer stage architecture, measurement and exposure cannot be performed at the same time. Therefore, the maximum throughput is only about 25 wafers per hour. The i-MAT alignment system currently only reads a small number of alignment marks at the wafer edge and cannot analyze full-wafer errors simultaneously like ASML's Twinscan architecture.
With the above inherent defects, even if potential customers such as Kioxia and Micron express interest in NIL, their speeches at technical seminars and public occasions are still conservative. In addition, representatives from many memory manufacturers bluntly stated that congenital defects are still the biggest challenge, and the cost and lifespan of templates limit the feasibility of production. The overall view in the industry is that it will be difficult for NIL to enter mainstream advanced process nodes in the short term.
Overall, from a theoretical perspective, NIL combines the advantages that the semiconductor industry dreams of, including high resolution, low energy consumption, low cost, and no random photon errors. However, if the problems of photomask mechanical strength and defect replication cannot be fundamentally solved, this technology may remain at the laboratory level.
In addition, as industry insiders describe it, NIL is like a perfectly designed precision clock, with performance and cost far exceeding those of competing products. But the key gear is made of glass. Therefore, it seems perfect, but it cannot survive actual operation.. Currently, Canon continues to invest in research and development and is trying to use NIL technology to develop mid-level process applications such as memory and display driver ICs. If the mask durability and mass production stability can be gradually improved, NIL will still have the opportunity to gain a foothold in specific application markets. However, in the short term, ASML’s EUV hegemony is still difficult to shake, and NIL’s disruptive potential still needs more time to prove.