Search

Extreme ultraviolet lithography (EUVL) has been adopted into high volume production for advanced logic device manufacturing. Due to the continuous size scaling requirement for interconnect fabrication, EUVL with self-aligned double patterning (SADP) formation has attracted substantial research attention. The current challenge in EUV SADP is the pattern transfer process from lithography to mandrel formation. In this step, the target critical dimension (CD) of the feature needs to shrink by half from the lithography CD during the etch process. The increasing aspect ratio during this etch potentially deteriorates the pattern validity and the line edge roughness (LER). In addition to these challenges, EUVL has a fundamental bottleneck due to stochastic effects, which can lead to device degradation by defect formation and edge placement error (EPE). LER of the line and space pattern is one of the main contributors to EPE. Effective methods of LER reduction in both process and integration are needed in order to reduce pattern variation and boost device performance. In our study, we examine a technique to reduce LER on the EUV SADP line pattern. This technique involves the surface modification on the spin-on carbon (SOC) layer in the patterning stack and tone inversion process. We had found a trend between surface hydrophobicity of the SOC and the EUV SADP LER performance. The condition that increased the hydrophobicity of the SOC resulted in a lower LER performance after tone inversion. The tested conditions include direct current superposition (DCS) function with H2 plasma, fluorocarbon plasma, and the combination of DCS with H2 plasma and trimethylsilane dimethylamine deposition. On 20-nm pitch EUV SADP, this technique shows 26% of LER improvement from lithography to SADP formation. PSD analysis recorded about 6% and 30% of the LER improvement at the correlation length of >200 nm and 200 to 30 nm, respectively. A demonstration of this technique for a further scaling to 15-nm pitch also shows an LER reduction of 30% from lithography to SADP formation.

Extreme ultraviolet lithography (EUVL) has been adopted into high volume production for advanced logic device manufacturing. Due to the continuous size scaling requirement for interconnect fabrication, EUVL with self-aligned double patterning (SADP) formation has attracted substantial research attention [1]–[6]. Double patterning techniques in EUVL achieve pitch halving in the final feature by using the spacer defined approach and self-aligned block (SAB) mitigates the block placement error by leveraging etch selectivities and material filling capability. The current challenge in EUV SADP is the pattern transfer process from lithography to mandrel formation. In this step, the target critical dimension (CD) of the feature needs to shrink by half from the lithography CD during the etch process. The increasing aspect ratio during this etch potentially deteriorates the pattern validity and the line edge roughness (LER) [5]. In addition to these challenges, EUVL has a fundamental bottleneck due to stochastic effects which can lead to device degradation by defect formation and edge-placement-error (EPE) [7]–[10]. LER of the line and space pattern is one of the main contributors to EPE. Effective methods of LER reduction in both process and integration are needed in order to reduce pattern variation and boost device performance. In our research, we examine three approaches to reduce LER on the EUV SADP line pattern. This includes photoresist surface smoothing techniques, patterning layer material study, and tone inversion integration. The photoresist surface smoothing techniques involve a specific plasma process on the EUV chemical amplified resist (CAR) to achieve > 15% of improvement on LER from lithography to post etch performance. The patterning layer material study reveals an optimum patterning stack to minimize etch-induced line wiggling and etch selectivity requirements for LER performance. Finally, a first demonstration of EUV SADP tone inversion process integration is presented as a method to provide additional benefits to LER reduction. A detailed analysis of line performance from each processing step will be examined.

As future patterning processes reach the limit of lithographic printability, continuous innovation in mandrel trim or shrink strategies are required to reach sub-20 nm line-space patterning. Growing concerns of lithography defectivity, mask selectivity, line edge roughness (LER), line width roughness (LWR), and critical dimension uniformity (CDU) present significant challenges towards this goal. The authors compare various alternative mandrel trim strategies to highlight potential solutions and drawbacks towards enabling successful printing of mandrels used in extreme ultraviolet (EUV) multi-patterning schemes. Through this comparison, the authors demonstrate the challenges of maintaining adequate pattern transferability while keeping aspect ratio-driven line roughness and material selectivity under control. By process partitioning, the limitations of traditional lithography and etch trimming strategies are highlighted, suggesting the need for new methods of CD reduction after the pattern has been transferred. These new trimming methods offer flexibility in CD control without negatively impacting the mandrel profile and demonstrates better tunability across different material sets, allowing for evaluation of different mask and mandrel material combinations for downstream process optimization.

For logic nodes of 7 nm and beyond, back-end-of-line (BEOL) trench patterns have a critical pitch of less than 40 nm, directly affecting the plasma etch process window of the dual damascene etch process. Feature size dependent etch depth (reactive ion etch, RIE lag), hard mask selectivity, and ultra-low-k (ULK) damage have become significant challenges that must be overcome in order to meet target device performance. Recently, atomic layer etching has been used to widen the plasma etch process window in terms of selectivity and process control [S. Sherpa, P. L. F. Ventzek, and A. Ranjan, J. Vac. Sci. Technol. A 35, 05C310 (2017); T. Tsutsumi, H. Kondo, M. Hori, M. Zaitsu, A. Kobayashi, T. Nozawa, and N. Kobayashi, J. Vac. Sci. Technol. A 35, 01A103 (2017)]. In this work, the impact of a quasiatomic layer etch (QALE) process, a conventional continuous wave plasma, and a pulsed plasma process on ULK materials were investigated to determine the benefits of an ALE process approach for BEOL etching. Both blanket ULK film and patterned ULK samples were used for this study. The ULK etch damage from each process was characterized using Fourier transform infrared spectroscopy and x-ray photoelectron spectroscopy on three different ULK films. From patterned samples, it was determined that QALE could be used to successfully suppress RIE lag in low-k materials at advanced pitches, while keeping low-k damage to a minimum. In addition, the QALE technique showed improved hard mask selectivity and resulted in lower line edge pattern roughness. Based on this study, the authors concluded that QALE is a powerful plasma etch method to overcome BEOL etch challenges at advanced pitches.For logic nodes of 7 nm and beyond, back-end-of-line (BEOL) trench patterns have a critical pitch of less than 40 nm, directly affecting the plasma etch process window of the dual damascene etch process. Feature size dependent etch depth (reactive ion etch, RIE lag), hard mask selectivity, and ultra-low-k (ULK) damage have become significant challenges that must be overcome in order to meet target device performance. Recently, atomic layer etching has been used to widen the plasma etch process window in terms of selectivity and process control [S. Sherpa, P. L. F. Ventzek, and A. Ranjan, J. Vac. Sci. Technol. A 35, 05C310 (2017); T. Tsutsumi, H. Kondo, M. Hori, M. Zaitsu, A. Kobayashi, T. Nozawa, and N. Kobayashi, J. Vac. Sci. Technol. A 35, 01A103 (2017)]. In this work, the impact of a quasiatomic layer etch (QALE) process, a conventional continuous wave plasma, and a pulsed plasma process on ULK materials were investigated to determine the benefits of an ALE process approach for BEOL etching. Both blank...

Atomic layer etching (ALE) of SiO2 was studied by alternating exposure of a 5 nm-thick SiO2 film on Si substrate to (1) a plasma beam emanating from a c-C4F8 inductively coupled plasma (ICP), to grow a fluorocarbon (FC) film composed mainly of CF2, and (2) an energetic (130 eV) Ar+ ion beam extracted from a separate Ar ICP. In situ x-ray photoelectron spectroscopy was used to analyze the chemical composition of the near-surface region, and to quantify the thickness of the FC and SiO2 films. A very thin (3–6 A), near self-limiting thickness CF2-rich FC film was found to deposit on the SiO2 surface with exposure to continuous or pulsed power C4F8 plasma beams, under conditions that generated a large relative flux of CF2. Following this, a FC film of similar composition grew at ~10 times slower rate. Exposure of the thin film to the Ar+ beam led to removal of 1.9 A SiO2. An estimated yield of 1.3 SiO2 molecules-per-Ar+ was found for a single ALE step. The rate of 1.9 A/cycle persisted over multiple ALE cycles, but a carbon-rich residual film did build up. This film can be removed by a brief exposure to an O2-containing plasma beam.

Silicon nitride (SiN, where Si:N ≠ 1:1) films low pressure-chemical vapor deposited on Si substrates, Si films on Ge on Si substrates, and p-Si samples were exposed to plasma beams emanating from CH3F/O2 or CH3F/CO2 inductively coupled plasmas. Conditions within the plasma beam source were maintained at power of 300 W (1.9 W/cm3), pressure of 10 mTorr, and total gas flow rate of 10 sccm. X-ray photoelectron spectroscopy was used to determine the thicknesses of Si/Ge in addition to hydrofluorocarbon polymer films formed at low %O2 or %CO2 addition on p-Si and SiN. Polymer film thickness decreased sharply as a function of increasing %O2 or %CO2 addition and dropped to monolayer thickness above the transition point (∼48% O2 or ∼75% CO2) at which the polymer etchants (O and F) number densities in the plasma increased abruptly. The C(1s) spectra for the polymer films deposited on p-Si substrates appeared similar to those on SiN. Spectroscopic ellipsometry was used to measure the thickness of SiN films etched u...

A CH3F/CO2 inductively coupled plasma (ICP), sustained in a compact plasma reactor, was investigated as a function of power (5–400 W) and feed gas composition, at a pressure of 10 mTorr, using optical emission spectroscopy and rare gas actinometry. Number densities of H, F, and O increased rapidly between 74% and 80% CO2, ascribed to the transition from polymer-covered to polymer-free reactor walls, similar to that found previously in CH3F/O2 ICPs at 48% O2. Below 40% O2 or CO2, relative emission intensity ratios were almost identical for most key species in CH3F/O2 and CH3F/CO2 ICPs except for higher OH/Xe (a qualitative measure of OH and H2O densities) over the full range of CH3F/O2 composition. The number density of H, F, and O increased with power in CH3F/CO2 (20%/80%) plasmas (polymer-free walls), reaching 4.0, 0.34, and 1.6 × 1013/cm3, respectively, at 300 W. The CO number density increased with power and was estimated, based on self-actinometry, to be 8.8 × 1013/cm3 at 300 W. The CO2 number density...

CH3F/O2 inductively coupled plasmas at 10 mTorr were investigated using optical emission spectroscopy. A 'self-actinometry' method was developed to measure the absolute number density of CO that formed in reactions following dissociation of CH3F and O2 in the plasma. In this method, small amounts of CO were added to the plasma, leading to small increases in the CO emission intensity. By carefully accounting for small perturbations to the plasma electron density and/or electron energy distribution, and by showing that very little of the CO added to the plasma was decomposed by electron impact or other reactions, it was possible to derive absolute number densities for the CO content of the plasma. With equal fractions (0.50) of CH3F and O2 in the feed gas, the CO mole fraction as a function of plasma power saturated at a value of 0.20–0.25. As O2 in the feed gas was varied at a constant power of 100 W, the CO mole fraction went through a maximum of about 0.25 near an O2 feed gas fraction of 0.5. The relative CO number densities determined by 'standard' actinometry followed the same functional dependence as the absolute mole fractions determined by self-actinometry, aided by the fact that electron temperature did not change appreciably with power or feed gas composition.