YORKTOWN HEIGHTS, N.Y. IBM Corp.'s T.J. Watson Research Center has announced that it is releasing its proprietary sub-100-nanometer lithographic mask repair technology for general license.
IBM uses the femtosecond-laser-based technology to repair masks damaged by metal splatter, gallium staining and pitting. The all-optical method also avoids problems created by ion beam methods, currently the main competing technology at sub-100-nm feature sizes. The charged particles used in ion beam methods tend to interact with the metal and quartz materials of a mask, making it difficult to control the process, whereas photons are electrically neutral, eliminating many headaches, the company said.
"IBM wants to license its mask repair technology, which we believe is key to the continual process of scaling down feature sizes on wafers. It's been a real enabling technology for IBM," said Richard Haight, an IBM researcher specializing in high-speed physics. Fellow researchers Alfred Wagner and Peter Longo co-developed the technology.
"There is a tremendous investment in mask manufacturing lithography is the way that all semiconductor manufacturers print chips, and it's going to continue that way. Consequently, there is a constant push to find ways to improve the technology and to reduce the cost and time it takes to produce essentially a perfect mask or perfect enough to make chips that will function properly," Haight said.
Fast laser burst
The IBM mask repair technology makes use of a 100-femtosecond laser burst that is so fast that its spatial resolution is not degraded by thermal diffusion. Basically, the longer the exposure time, the more the heated area enlarges as energy moves away from the region of interest. By keeping that time as short as possible, the resolution of the beam is only limited by optical diffraction.
The beam is focused down to a 150-nm diameter spot and the intensity increases from the circumference to the center with a Gaussian (bell-shaped) profile. By reducing the intensity of the beam, it is in fact possible to create a smaller thermal footprint of about 80 nm, since the intensity on the outer edge of the spot becomes too low to affect the mask material.
Any shaped defect can be removed by scanning it with a resolution of 80 nm, or in "nibble mode," mask features can be trimmed to a root mean square precision of less than 5 nm.
"The patterns on today's advanced lithographic masks have become too complex and expensive to throw away just because they have a defect, so in order for the whole lithographic method to continue scaling down, the tools used to repair masks must be scaled down too," Haight said.
With current lithographic methods, the features on a chip are usually four times smaller than those on its mask (4x reduction printing), but because features on the wafer are dipping below 100 nm, now it is difficult just to get the light through the features of the mask. Consequently, assist features are added to allow extra light to get through and properly expose the chip. Assist features on the mask can be just as small as the features on the wafer, necessitating IBM's sub-100-nm repair technology.
Lasers once were the best way to repair masks, but the nanosecond-laser tools projected a high-power beam onto a defect long enough to melt the chrome, thereby boiling away a defect, causing splatter. Also, the quartz substrate is heated in the process, resulting in thermal diffusion that limits the beam's spatial resolution to 500 nm.
"The problem with nanosecond lasers is that it's a thermal process that melts the defect . . . it boils and spits so that you end up splattering the chrome around, creating more defects. Also, the metal can diffuse into the glass quartz substrate of the mask, causing pitting," Haight said.
No ion beam heat
To solve the thermal problems, ion beams were adopted to repair the most advanced masks today. Ion beams blast the chrome off while it is still in the solid state by using a stream of gallium ions to sputter etch chrome defects off the mask. The chrome is then carried away by merely flowing a gas over the wafer.
However, IBM's own mask-making operations started running into pitting problems with ion beams, because unless the beam is switched off at exactly the right moment, it begins diffusing gallium into the quartz.
"The problem we found with ion beams is that when you get through the chrome, you begin implanting gallium, and that stains the glass quartz . . . You can also end up etching the glass, creating a pit . . . Also, since the mask is glass, you have charging problems that make it hard to precisely aim the beam," Haight said.
Focused ion beams, according to IBM, also require frequent, time-consuming setup and calibration, when compared with the ease of use for lasers. In addition, the small beam currents needed to attain good spatial resolution result in proportionally slower repair rates.
By using 100-femtosecond laser bursts, spaced one every thousandth of a second, IBM claims to have avoided heating the glass substrate. With this method, it takes over a picosecond for the molecules in the glass to convert the laser's energy into thermal heat, and its bursts are 10 times shorter (100 femtoseconds = 0.1 picoseconds).
"By going to a very short pulse we found that we could remove the metal without damaging the glass quartz in a nonthermal process . . . The pulse excites the metal so quickly that it goes directly from a solid to a plasma . . . Defects blast right off with very little debris, and any debris there is can be subsequently removed with the same laser," Haight said.
By avoiding thermal effects altogether, IBM claims it avoids all the pitfalls of previous laser methods, while getting even more reliable results than today's ion beams.
"For any thermal process you have thermal diffusion, which spreads out the whole effect, which is bad. Plus, because our technique is nonlinear, we can actually operate at optical resolutions below the diffraction limit," said Haight.
"We can now image and ablate at the same wavelength, so that we can see what we are doing as we do it. We have such good visibility, in fact, that our technicians use our tool to see the defects, because you can see smaller defects with our tool than with any of the other tools available. . . . They use our tool to repair all of their chrome defects," Haight said.
IBM specified a variety of metrics for its technology, including the basic 150-nm size of the beam itself. By controlling the nonlinear beam's intensity, however, the company has been able to make 80-nm small lines and 75-nm-diameter holes in chrome. And by nibbling away at the edge of a defect, IBM has successfully shaved off defects as small as 5 nm.
"On a day in and day out basis in our mask house, our technicians can now routinely make repairs with a resolution of about 150 nm. And we have demonstrated that if we are very careful about the intensity of the laser, we can make 80-nm repairs in the chrome," Haight said.
Since most mask features are four times bigger than features on the final chip, those sub-100-nm metrics also scale down. For instance, the basic 150-nm size of the beam when repairing masks corresponds to repairing 35-nm features on the wafer.
Likewise, the 80-nm lines made by backing off on laser intensity correspond to 20-nm features on the chip. And the 5-nm nibbles that can be sliced off mask defects correspond to slicing off 1.25-nm features from chips. Coincidentally, that's the same size as the diameter of a nanotube.
An audio recording of reporter R. Colin Johnson's full interview with Richard Haight can be found online at AmpCast.com/RColinJohnson.