An optimized process for dry stripping photomasks
05/01/2003
Overview
Increasingly, new photoresists and novel polymer-rich plasma etch processes used in photomask manufacturing require that conventional wet stripping must be replaced with dry plasma-stripping techniques. But these dry processes bring a whole new set of process concerns. This work optimized a microwave-plasma stripping process for the best photoresist-to-chrome oxide selectivity and photoresist etch rate, while maintaining antireflective coating integrity.
Organic polymers are an integral part of the photomask manufacturing flow. They serve as photoresists for chrome on glass (CoG) patterns, for phase-shift mask (PSM) levels, and as sidewall protection during plasma etch of several materials, such as MoSiON and SiO2 for attenuated and alternated PSMs.
The common method for stripping photoresists is wet stripping using amineous solvents and acids such as H2SO4-H2O2, which work well with PBS and Novolac resists. Newer e-beam sensitive photoresists, however, are designed for improved plasma etching resistance for application in chrome plasma etching; this makes them difficult to strip with established wet processes. In particular, sidewall polymers from plasma etching often show a high resistance to wet stripping chemistries.
Oxygen-plasma stripping is an alternative that addresses these problems. Here the simplest system is a microwave barrel reactor — the most economic system for low-throughput photomask applications. Microwave oxygen plasma attacks the chromium oxide (CrOx) antireflective coating used on masks, however, leading to remarkable loss of the antireflective coating after the photoresist has been stripped. A plasma chemistry modification is necessary to avoid this reaction; the addition of hydrogen (H2) to the plasma completely eliminates any etching of CrOx layers, which are crucial for today's steppers.
To prove the capability of this approach, we have performed a series of resist stripping experiments on binary 6 ¥ 6 in. ¥ 250 mil CoG masks. The absorber film was a 70nm Cr 33nm CrOx bilayer coated with either 570nm of iP3600 or 300nm ZEP7000 photoresist. Our plasma system was a semiautomated TePla 300 (a converted barrel ashing system) with a 1000W 2.45GHz microwave plasma generator. For the simplicity of the reactor setup, we relied on total gas flow (all gases were electronic grade) without any pressure control. We also estimated plasma surface temperature from pyrometric measurements taken from a thin alumina plate located inside the plasma.
We recorded plasma emission spectra with a Hamamatsu MPM 7460 optical multichannel spectrometer. To gather mask data, we measured film thickness with a Nanometrics Nanospec 6100; CD and registration metrology with a Leica LWM 250 i-line transmitted light and a Leica LMS IPRO reflected light measurement system; and defects with a KLA Starlight.
CrOx etch rate
Looking at the removal of CrOx by oxygen plasma, we know that microwave plasma ions see a very low bias voltage toward the substrate close to the plasma potential, which is in the range of 20V, so physical sputtering does not explain this effect. The reaction shows a time lag, indicating a need for a threshold temperature and a chemical etching mechanism.
X-ray photoelectron spectra (XPS) show that in the CrOx layer Cr is present in III or IV oxidation states, most likely as CrO2. This indicates that the etching reaction probably proceeds via further oxidation of the chromium oxide to CrO3. Although under normal conditions CrO3 dissociates at elevated temperatures, it probably has significant volatility under the given process conditions because of its covalent character, and it could play a dominant role in the plasma etching mechanism. The plasma emission spectra also show some CrO lines, further supporting this line of reasoning.
Chemical etching by formation of carbonyls, which would be volatile under these conditions, can be excluded because the etch reaction occurs even in the absence of carbon from photoresist in a thoroughly cleaned reactor.
With this analysis, we concluded that the addition of a gas with reducing properties — the simplest choice is H2 — would avoid the oxidation of CrOx. Indeed, we found that increasing H2 concentration in the reactive gas increased reaction lag time (Fig. 1) and reaction onset temperature shifts to higher values.
 Figure 1. CrOx etch lag time and etch onset temperature vs. H2 concentration in plasma (400 sccm total flow, 1000W MW-power). |
null
Resist etch rate
After finding a suitable process window, we set up a designed experiment (DoE) to optimize resist stripping rate and resist-CrOx selectivity. With the microwave generator at 1000W, the test was a simple two-level, two-factorial DoE with three centerpoints varying hydrogen concentration in the O2-H2 mixture from 20%–50% and total flow from 200–600 sccm. Without active pressure control, we simply accepted the pressure variation (i.e., 0.4–0.8mbar) resulting from MFC controlled flow variation. We set overetch to determine CrOx etch rate at 60 sec.
We found no significant CrOx etch rate difference at any of the parameter settings. Also, H2 concentration has no significant impact on the resist-stripping rate. With increasing total flow, and therefore pressure, the resist-stripping rate increased in the range of 50–75nm/min; this is obviously attributable to increasing plasma density with higher pressure.
From these results, we chose a baseline recipe using 1000W of microwave power, 300 sccm O2, 300 sccm H2, and a typical pressure of 0.75mbar. This maximized strip rate without attack on the CrOx layer.
Resist etch rate
Under the given conditions, resist stripping is also probably chemistry-driven and should be strongly temperature dependent, which we confirmed by plotting strip rate vs. maximum surface temperature in the plasma, giving roughly a 5nm/min increase with every 10°C increase in temperature.
 Figure 2. Substrate surface temperature and CO emission line intensity vs. etch time. |
null
Due to the large heat capacity of the quartz glass body of a photomask, substrate temperature slowly increases during the initial phase of the stripping process. Substrate heating is mostly generated by recombination of reactive radicals at the substrate surface and the reaction enthalpy of the stripping process itself. Figure 2 shows the correlation of temperature increase over plasma time with the intensity of the CO plasma emission line at 519nm. As shown before, also in the initial stage, there is a correlation of the photoresist stripping rate and the substrate surface temperature.
As the resist stripping reaction is driven by chemical etching, there should be a strong decrease of stripping rate with increasing total amount of resist. This "macroloading effect" is determined by the depletion of reactive species at the surface reaction. To check the consequences of this effect on process time, we stripped a series of masks with different photoresist loads (Fig. 3).
 Figure 3. Resist stripping rate vs. total photoresist load on a mask. |
null
These data clearly show the need for endpoint detection so that throughput can be optimized. A typical optical plasma emission spectrum has prominent H (e.g., 434, 486, 656nm) and O (e.g., 777, 845nm) lines and very weak CO emission band lines (e.g., 450, 482, 519, 560nm).
 Figure 4. Intensity traces of H, O and CO plasma emission line during the strip process. |
null
We found that the most intense H emission line at the 656nm wavelength gives a weak intensity change when reaching the strip endpoint (Fig. 4), although the mechanism for this decrease is not yet fully understood. (Possibly after reaching endpoint, oxygen is then available for reaction with hydrogen.) The O line at 777nm shows a slight intensity increase after reaching endpoint, but also no clear endpoint signal. Fortunately, despite very low intensity, the CO signal trace at 519nm shows the best signal-to-noise ratio of all signals and a clear endpoint signature. Thus, we set up an endpoint detection system using a high transmission interference filter for the 519nm wavelength of the CO line and a high-sensitivity photodiode as detector.
Chromium film, quartz substrate integrity
The very low CrOx etch rate indicates that the process should not result in any damage to the Cr-CrOx film on a mask. We checked this with an experimental mask having a 132 ¥ 132mm patterned area, which we pre-measured for CDs and defects. We coated this mask with 570nm of iP3600 resist and plasma stripped the resist using our standard recipe plus a 2.5 min overetch, then repeated the measurements. Defect measurements confirmed that there was no damage to the CrOx layers. CD measurements showed deviations within the range of measurement error for the tools used (see table).
 |
null
Placement of mask structures is crucially important and can be affected by plasma processing, when heating of the mask can exceed 100°C; registration error caused by thermal expansion must be completely reversible. To check this, we pre-measured mask registration, exposed these masks to a plasma for 10 min, and then remeasured the registration after varying waiting times. We wet-stripped some masks as a reference. We found no significant change in registration due to the plasma stripping process, with data variation within the range of measurement repeatability (Fig. 5).
 Figure 5. Registration change after plasma and wet strip. |
null
Applications
 Figure 6. Chromium line after 0.5 min overetch (left), 3.5 min overetch (center), and 3.5 min overetch plus wet clean (right). |
null
Finally, we studied the applicability of our process to stripping ZEP 7000 e-beam resist. We etched several masks with a typical product design (i.e., 50% clearfield) until endpoint, varying overetch from 0.5–3.5 min. Analysis with SEM showed that in all cases there were residues left on the chrome structures. The amount of residue decreased with overetch time, but even at 3.5 min (130%) overetch these residues could not be removed completely. A subsequent Piranha wet clean was necessary to clean the chrome surface totally (Fig. 6). The Piranha clean is necessary to remove inorganic particles but — in combination with the plasma stripping — avoids the use of harmful and expensive amineous solvents.
 Figure 7. Chrome line with sidewall polymers after wet strip (left), and the same line after combined plasma and wet treatment (right). |
null
For the fabrication of alternated PSMs, trenches have to be plasma-etched into a quartz mask substrate. To form vertical sidewall profiles, fluorocarbon etch gases are used to build up sidewall protecting polymers. These polymers are Teflon-like and therefore hardly strippable with wet chemical processes. To solve this problem, we used our combined plasma and wet strip processing to remove the sidewall polymers completely (Fig. 7).
Guenther Ruhl, Pavel Nesladek, Infineon Technologies AG, Munich, Germany
Astrid Boesl, PVA TePla AG, Feldkirchen, Germany
Acknowledgments
The authors thank Andreas Niederhofer for his help with registration analysis, and Frank Scharberth for SEM images; both are with Infineon Technologies Mask House.
For more information, contact Guenther Ruhl, senior engineer, dry etch, at Infineon Technologies AG Mask House, Balanstrasse 73, D-81541 Munich, Germany; ph 49/89-234-23171, fax 49/89-234-9552408, [email protected]. Pavel Nesladek is a process engineer, dry etch, at Infineon Technologies AG Mask House. Astrid Boesl is an application engineer at PVA TePla AG.