If a picture is worth a thousand words, how many words is a photomask worth? SPIE’s 2017 Advanced Lithography Conference is underway and Dr. Frank Abboud, Vice President Technology Manufacturing Group and General Manager, Intel Mask Operation kicked off his keynote presentation “Photomask Technology Challenges for Upcoming technology Nodes” by comparing the amount of digitized data in a 14nm photomask to that in a two hour HD movie. A six inch square 14nm photomask contains more images than ten HD movies…that’s a lot of pictures! Loading this photomask onto a cutting-edge immersion scanner creates the world’s fastest form of data transfer, outrunning other methods of data transfer from 5G to optical links by four orders of magnitude. This ability to transfer massive amounts of data quickly to the wafer has been an important enabler of Moore’s Law for the past five decades and will continue to be so for the foreseeable future.
Abboud progressed to explain the evolution of photomask technology, from contact printing up through extreme ultraviolet lithography (EUVL). Contact printing and later proximity printing enjoyed mainstream use in the 1960’s and 1970’s. 1X projection masks took over proximity printing masks due to proximity’s inability to keep masks defect free. This theme of zero defect masks will prove to be a persistent one for years to come. Maskmaking blossomed in the 1980’s with over 200 writers capable of printing photomasks with feature sizes equal to those on the wafer. Many companies developed e-beam writers to support this market. The mid 1980’s saw the introduction of 5x reduction lithography steppers: this provided a ten year “holiday” in resolution, but the smaller number of die per mask meant that zero defect masks were “must have”.
The maskmaker’s holiday ended in the late 1990’s, when lithographers began printing features smaller than the wavelength of the 193nm light in the scanner. This motivated several new mask technologies. Phase shift masks (PSM) became common, and required writing and etching multiple layers on a photomask. Abboud shared the insight that laser based mask writers had existed since the mid 1980’s as fast, loose layer tools, but their ability to write patterns without charging up the mask proved to be a PSM technology enabler. On the e-beam side, stringent pattern fidelity and resolution enhancements drove the transition from 10keV to 50keV e-beam, and 50keV drove introduction of chemically amplified resists into the mask shop to enable faster write times with less heating of the mask during writing.
In the 2000’s, resolution enhancement techniques (RET) and optical proximity correction (OPC), introduced in the 1990s, became increasingly complex. Today, the pattern on an inverse lithography technique (ILT) mask bears no resemblance to the pattern printed on the wafer and the mask has become a diffractive optical element, altering the intensity, phase, polarization, and direction of the light going through it. Simply inspecting the pattern on the mask using existing tools is no longer a good predictor of whether the mask is defect free, and aerial image-based inspection, where the inspection tool emulates a lithography scanner to make this judgement, became a necessity.
Abboud explained that the amount of data required to write each mask and the total number of masks has increased, creating productivity challenges as each mask can take over two days to write vs. the old norm of less than a half day. Abboud shared that in 2012 Intel and the industry predicted this growing data volume and maskmaking productivity challenge. The mask industry had also consolidated due to high capital costs and the benefits of mask/wafer co-optimization. This combined with the realization that slower photoresists produced better pattern fidelity, drove the need for development of multibeam mask writing (MBMW) and creation of the MBMW Consortium with industry partners, which enabled IMS to develop and be first to market with MBMW capability. This has returned mask write times to the half-day norm even when using slow, high resolution resist.
Abboud finished his talk by explaining the impact of the transition to EUVL on maskmaking, stating that EUVL masks will be ready and able to continue to extend Moore’s Law. He explained that EUV masks do the same job as optical masks, although making EUV masks impacts many of the process steps in the mask shop, starting at film on the bottom of the mask blank up through the final pellicle inspection. Mask equipment and blank suppliers have developed novel tools and materials to enable this transition with the help of industry consortia. Even with these broad changes, Intel’s mask shop is shipping masks to meet our development needs, and EUV masks are making progress toward achieving the goal of zero printable defects. In closing, Abboud stated that mask innovations will continue to enable the extension of Moore’s Law and future semiconductor industry growth, and the five decade long trend will continue.
Janice Golda is director of Lithography Strategic Sourcing in the Technology and Manufacturing Group at Intel.