We report a protocol for combining the atomic metrology of the Scanning Tunneling Microscope for surface patterning with selective Atomic Layer Deposition and Reactive Ion Etching. Using a robust process involving numerous atmospheric exposures and transport, 3D nanostructures with atomic metrology are fabricated.
Reducing the scale of etched nanostructures below the 10 nm range eventually will require an atomic scale understanding of the entire fabrication process being used in order to maintain exquisite control over both feature size and feature density. Here, we demonstrate a method for tracking atomically resolved and controlled structures from initial template definition through final nanostructure metrology, opening up a pathway for top-down atomic control over nanofabrication. Hydrogen depassivation lithography is the first step of the nanoscale fabrication process followed by selective atomic layer deposition of up to 2.8 nm of titania to make a nanoscale etch mask. Contrast with the background is shown, indicating different mechanisms for growth on the desired patterns and on the H passivated background. The patterns are then transferred into the bulk using reactive ion etching to form 20 nm tall nanostructures with linewidths down to ~6 nm. To illustrate the limitations of this process, arrays of holes and lines are fabricated. The various nanofabrication process steps are performed at disparate locations, so process integration is discussed. Related issues are discussed including using fiducial marks for finding nanostructures on a macroscopic sample and protecting the chemically reactive patterned Si(100)-H surface against degradation due to atmospheric exposure.
As nanotechnology becomes more important in a wide variety of arenas, understanding the structures being formed gains importance, especially in fields of lithography and electronics. To emphasize the importance of metrology at the nanoscale, specifically at scales below 10 nm, it should be pointed out that a variation in feature size of only 1 nm indicates a fractional variation at least 10%. This variation can have significant implications for device performance and material character.1,2–4 Using synthetic methods, very precisely formed individual features such as quantum dots or other complex molecules can be fabricated,2,5,6 but generally lacking the same precision in feature placement and orientation, despite work toward improving size and placement control. This paper demonstrates an approach for fabricating nanostructures with near atomic size precision and atomic precision in feature placement, as well as with atomic metrology in feature placement. Using the atomic precision of Scanning Tunneling Microscope (STM) induced Hydrogen Depassivation Lithography (HDL), atomically precise patterns with chemically sensitive contrast are formed on a surface. Selective Atomic Layer Deposition (ALD) then applies a hard oxide material in the patterned areas, with Reactive Ion Etching (RIE) ultimately transferring the patterns into the bulk material, as shown schematically in Figure 1. Combining the highly precise HDL process with the standard ALD and RIE processes results in a flexible method to produce nanostructures on a surface with arbitrary shape and positioning.
Figure 1. Primary Nanofabrication Process Steps. As an example, a 200 nm x 200 nm square is shown. Each circled arrow indicates a step of atmospheric exposure and transport between sites. After UHV sample prep, the sample is patterned using UHV HDL followed by STM metrology (top left). ALD is then performed, followed by AFM metrology (right). RIE transfers the patterns into Si(100), followed by SEM metrology (bottom left). Please click here to view a larger version of this figure.
The most precise lithography to date usually involves scanned probe techniques, specifically STM-based patterning where atomic resolution patterning and functionalization has been demonstrated for many applications.7 Previously, atom manipulation has produced nanostructures with ultimate precision by using individual atoms as building blocks,8,9,10 but the nanostructures required cryogenic conditions and thus lacked long-term robustness. RT atom manipulation by removal of hydrogen atoms from the surface has been shown, specifically HDL.11,12,13 HDL promises to enable new classes of electronic and other devices based on the spatial localization of surface contrast. Using HDL without further processing, various device architectures are possible including dangling bond wires or logic devices.14,15,16 In addition to providing electrical contrast, HDL can introduce chemical contrast on the surface where the passivating H layer has been removed, in effect creating a template for further chemical modification. This chemical modification has been demonstrated on silicon and other surfaces, showing selectivity for deposition of metals,17 insulators,18 and even semiconductors.16,19 Each of these examples produces two dimensional structures, so other processing steps must be used to produce true three dimensional structures with the atomically resolved control promised by HDL. Previously, this has required repeated patterning,19,20,21 annealing,22 or less well resolved processes such as tip-based e-beam induced deposition.23
Similar to e-beam lithography, HDL uses a localized flux of electrons to expose a resist. Several similarities exist such as the capacity to perform multi-mode lithography with variable spot size and patterning efficiency.24 However, the true power of HDL arises from how it differs from e-beam lithography. First, the resist in HDL is a monolayer of atomic hydrogen so that resist exposure becomes a digital process; the resist atom either is or is not present.25 Since the H atom placement corresponds to the underlying Si(100) lattice the HDL process can be an atomically precise process, although it should be noted that in this paper the HDL has nanometer precision as opposed to having atomic perfection and thus is not digital in this case. Since the electron source in HDL is local to the surface, the various modes of STM operation facilitate both throughput optimization as well as error checking. At tip-sample biases below ~4.5 V, lithography may be performed at the single atom level with atomic precision, known as Atomically Precise mode (AP mode). In contrast, at biases above ~7 V, electrons are emitted directly from the tip to the sample with wide linewidths and high depassivation efficiencies, known here as Field Emission mode (FE mode). HDL throughputs can then be optimized by careful combination of these two modes, although the overall throughputs remain small relative to e-beam lithography with patterning up to 1 μm2/minute possible. When the bias is reversed so that the sample is held at ~-2.25 V, electrons tunnel from the sample to the tip with extremely low depassivation efficiency, thus permitting inspection of the atomic structure of the surface both for error correction and for atomic scale metrology.
This nanostructure fabrication process shown in Figure 1 starts with an UHV-HDL step, as described above. Following HDL, the sample is vented to atmosphere, at which time the patterned areas become saturated with water, forming a thin (i.e., ~1 monolayer) SiO2 layer.26 After transport, the sample is inserted into an ALD chamber for deposition of titania (TiO2), with thicknesses around 2-3 nm deposited here, as measured by AFM and XPS.27 Since the titania reaction depends upon a water saturation of the surface, this process is possible despite atmosphere exposure which saturates the surface with water. Next, to transfer the ALD mask pattern into the bulk the sample was etched using RIE so that 20 nm of Si is removed, with the etch depth determined by AFM and SEM. In order to facilitate metrology steps, an Si(100) wafer is patterned with a grid of lines which are designed to be visible after UHV preparation by a long working distance optical microscope, AFM plan-view optical imaging, and low-magnification plan-view SEM imaging. To help identify the nanoscale structures, 1 μm2 serpentine patterns (serps) are patterned onto the samples with the most isolated nanopatterns located at fixed locations relative to the serps.
This combination of HDL, selective ALD, and RIE can be an important process for nanostructure fabrication, and it includes an atomic scale metrology as a natural byproduct of the process. Below, we include a detailed description of the steps involved to fabricate sub-10 nm nanostructures in Si(100) using HDL, selective ALD, and RIE. It is assumed that one is skilled in each of these processes, but information will be included related to how to integrate the various processes. Particular emphasis will be given to those unexpected difficulties experienced by the authors in order to prevent the same difficulties, especially related to transport and metrology.
1. Ex-Situ Sample Preparation
2. UHV Sample Preparation
3. Scanning Tunneling Microscopy and Lithography
4. Sample Transport
5. Atomic Layer Deposition
6. Atomic Force Microscopy (AFM)
7. Reactive Ion Etching
8. Scanning Electron Microscopy (SEM)
In the cases described here, HDL is performed using multi-mode lithography.24 In FE mode, performed with 8 V sample bias, 1 nA, and 0.2 mC/cm (equivalent to 50 nm/sec tip speed), the tip moves over the surface either parallel or perpendicular to the Si lattice, producing lines of depassivation. While this lineshape is very tip dependent, in the case here, the completely depassivated portion of the lines was approximately 6 nm wide, with tails of partial depassivation extending another 2 nm on either side of the line. When highly precise patterns are desired, AP mode lithography is performed using 4 V sample bias, 4 nA, and 4 mC/cm (equivalent to 10 nm/sec tip speed). The extent of the AP mode component of each pattern depends upon the width of the partially depassivated patterns produced using FE mode. See Figure 2 for examples of STM images of patterns on Si(100)-H for the various HDL modes. Figure 2A shows a small pattern produced using only AP mode HDL. Figure 2B is an example of a pattern written using multi-mode lithography, where the FE mode lines were approximately 6 nm wide, but were written on an 10 nm pitch, with approximately 2 nm of each edge written using AP mode HDL. The FE mode portions in the interior of the pattern were written on a pitch of 10 nm, so there are narrow regions within the pattern where HDL was incomplete. For large, imprecise patterns FE mode can be used alone, as in Figure 2C where an approximately 1 μm2 serp pattern was written on a 20 nm pitch.
Figure 2. Representative HDL patterns. (A) STM image of an HDL pattern written with AP mode lithography of 4V, 4 nA, and 4 mC/cm (10 nm/sec). (B) STM image of a multi-mode HDL pattern written using a combination of AP mode and FE mode (8V, 1 nA, 0.2 mC/cm). The FE mode line pitch was chosen to be slightly larger than the written linewidth to improve the visibility of the vectors used in writing. (C) FE mode lithography of a large locator serp written on a 20 nm pitch. Please click here to view a larger version of this figure.
In order to achieve the best mask production using atomically precise HDL patterns, a high degree of selectivity must be possible. Previously, the ALD selectivity has been investigated by XPS and other methods comparing deposition on unpatterned Si(100)-H and Si(100)-SiOx surfaces as an analogs for the unpatterned and patterned areas, respectively.27,31 Using atomic force microscopy (AFM), we observe similar results, as shown in Figure 3. 200 nm x 200 nm squares were patterned onto Si(100)-H with >90% of the hydrogen removed within the patterns, and then titania was deposited on the patterns. AFM analysis of the patterns compared to the background showed that there was a taller layer of growth on-pattern than the height of tallest defect on the background. Extrapolation of the growth curves for the on-pattern and background deposition indicate an identical growth rate, but with an incubation of 20 cycles for the tallest background growth. It is worth reiterating here that most of the background growth occurs with a much longer incubation time. As described above, the Ar purge step is of great importance in order to achieve selectivity because it helps remove physisorbed precursors. Preliminary results with other titania precursors, such as tetrakis(dimethylamino)titanium, show more background reactivity, presumably because the precursors physisorb too strongly to the Si(100)-H surface. To prevent background adsorption, careful attention to the precursor chemistry is of great importance.
Figure 3. Selectivity of deposition. (A) Sample AFM image showing TiO2 deposition on patterned and background areas. Deposition was carried out at 100 °C. (B) Deposition depths for various numbers of cycles. The crosses represent the height as measured by AFM of the growth “on pattern” relative to the background. The open circles show the height as measured by AFM of the tallest background deposition within an area of 200 nm x 200 nm near a patterned area. Please click here to view a larger version of this figure.
Given that it is possible to deposit onto HDL patterns, investigation of the limits of pattern features should be examined. While it has already been shown that the ALD produces broadened patterns relative to the HDL patterns, and that the etched structures are slightly shrunk relative to the masks, the effect of producing highly dense arrays still remains somewhat unresolved. Figure 4 shows the HDL, titania mask, and etched structures for an array of squares fabricated using FE mode HDL lines written at a pitch of 15 nm. In Figure 4A, the HDL pattern shows two serps—one rotated by 90 degrees relative to the other—written with an 8 V tip-sample bias, 1 nA current, and 0.2 mC/cm dose (or 50 nm/sec tip speed). There are clearly openings in the pattern of varying sizes. Within the openings themselves, some HDL has occurred, but it remains low—on the order of 20% H removal. Figure 4B shows an AFM image of the same pattern after mask deposition. Due to tip convolution effects, the openings in the pattern are difficult to resolve. However, a clear order is observable. Figure 4C is an SEM image of the same pattern after RIE. Approximately 60% of the desired openings were indeed transferred into the substrate, indicating that this pattern size and density is approximately the limit for effective nanostructure fabrication using FE mode HDL.
Figure 4. Array of openings. (A) STM of HDL with lines written using FE mode. Two serpentine patterns, rotated at 90 degrees relative to each other, are written with a pitch of 10 nm. (B) AFM image after 2.8 nm of ALD of TiO2 of the same pattern. (C) SEM of “hole” array after RIE to remove 20 nm of Si. Notice that some “holes” have failed to etch. Please click here to view a larger version of this figure.
Performing metrology on the nanostructures described above requires the ability to bridge the tip positioning during HDL and pattern location using other tools such as AFM and SEM. In contrast to other well-developed patterning tools with high-resolution position encoding such as e-beam lithography, the HDL performed here was performed with an STM without well controlled coarse positioning, so extra position identification protocols were used, as shown in Figure 3. First, a long-focal-length microscope is positioned outside the UHV system approximately 20 cm from the tip-sample junction. The sample is patterned with a square grid of 10 μm wide lines, 1 μm deep, on a pitch of 500 μm to facilitate identification of the tip location on the surface.
Figure 5. Pattern location images of sample. (A) Optical image of STM tip (left) and its reflection (right) in the Si(100) surface on an area of the sample with 500 μm pitch line pattern. The lines are 1 μm deep and 10 μm wide prior to UHV processing. Guide lines are included to show the line directions. (B) Close-up, de-skewed optical image of the tip (lower left) and its reflection (upper right). The centerpoint location between the tip and its reflection is identified within the 500 μm x 500 μm fiducial square. C: Closeup of patterning location with a 50 μm spot included for scale. D: 5 μm x 5 μm AFM image of an entire patterned area after ALD. E: 1 μm x 1 μm SEM image of one of the locator patterns after RIE. Please click here to view a larger version of this figure.
The first step to locating nanostructures is identifying the tip location on the surface,32,33 which is accomplished in this case using a long working distance microscope. Figure 5A shows an optical image of the tip when engaged with the sample, with dotted lines added to guide the reader for the directions of the fiducial grid. To locate the tip/sample junction the optical image is unskewed to make a square grid, as shown in Figure 5B, although there are errors in high temperature processing of the samples due to significant surface atom migration. This reduces the depth and visibility of the fiducial grid as imaged here, increasing the uncertainty in the tip position.32 While it has been shown previously that high temperature sample processing will induce significant atomic scale surface reconstruction, the grid pitch used here is great enough to have little effect on the surface reconstruction in the middle of the squared defined by the grid.34 However, near the edges of the patterned areas, step bunching does occur with asymmetry that depends upon the direction of current flow during sample preparation. 34 Since the optical imaging is performed at an oblique angle relative to the surface, small changes in height on one side of a trench relative to the other will induce additional uncertainty in pattern location—especially when compared to plan-view imaging as in AFM or normal SEM. After the tip engages the sample, the 10 μm focal spot size of the microscope coupled with the ~20 μm post-processing fiducial linewidth results in an approximate uncertainty in pattern position identification of ±27 μm. This defines the search window for using various techniques for pattern identification.
To facilitate the location of the smaller 10–100 nm features, additional large serps are added near the nanoscale patterns, as shown in Figure 5B. These 800 nm x 800 nm serps are written using FE mode HDL with vertical lines and gaps of 15 nm each. By aligning the AFM fast scan direction to be perpendicular to the serp lines (i.e., horizontal scanning), these patterns tend to show a high contrast in the AFM phase image due to the high spatial frequency of the topography, further facilitating pattern location. Once these patterns are found, it becomes much easier to find the smaller nanoscale patterns which are placed with approximately 100 nm accuracy relative to the large patterns.
For this nanostructure fabrication process, the sample undergoes atmospheric exposure between each major process step once HDL has been performed, as shown schematically in Figure 1. Given this, it must be assured that the sample does not degrade at any point in the handling. As shown above, there is a finite amount of background deposition during ALD, which is assumed to seed on background defect sites.31 Thus, improper handling such as extended atmosphere exposure can increase the number of background defects and reduce the apparent ALD selectivity. An additional surface degradation mechanism can occur during venting of the sample from the UHV load-lock to atmospheric conditions.29 To alleviate this problem, a spring-loaded sapphire chip which was mounted onto a linear actuator in UHV makes contact to the 125 μm thick sample mounting foil which contacts the sample to prevent surface degradation. Once the sample is at atmospheric conditions, the rate of dangling bond accumulation remains low (i.e., <0.1%/hr) for at least several hours, so as long as the sample is inserted into a stable environment such as ultra-pure Ar within less than 1 hr, the additional background deposition due to surface damage should remain low. At this point, it should be noted that the sample should not be stored in a vacuum environment, as this requires an additional vent/pump-down cycle, adding to the possibility of surface damage. This time between HDL and ALD is the point at which the sample is most sensitive since the etch mask has not yet been applied. After ALD, the sample still needs protection, but only to prevent additional mask growth due to silicon dioxide formation, a comparatively slow process.
In the patterns shown in Figure 4, the HDL removed >80% of the background H within the center of the patterns, with a spatial roll-off in the efficiency of the depassivation as the edge of the line is reached.24 Given the limits of very limited ALD on the background and incubation free growth on fully depassivated patterns (Figure 3), the edges of FE mode patterns where there is a transition from fully effective HDL and no HDL, show a transition of effectiveness of ALD mask growth. Below 70% H removal during HDL is where this transition starts to occur, indicating an approximate region of ~2 nm on each side of an FE mode line where partial mask deposition occurs.35 Additionally, ALD growth occurs in a “mushroom” manner,36 further broadening the mask relative to the HDL patterns so that a mask of 2.8 nm broadens any mask features by that amount. To summarize, the ALD linewidth can be expressed as Wm = Wsat + f(δH) + M where Wm is the total width, Wsat is the width of the line where the HDL has saturated to remove >70% of the surface H, f(δH) is the additional width due to the growth at each point due to the density of H remaining on the surface, and M is the additional linewidth due to the mushrooming of growth. δH depends upon the spatial distance from the saturated edge of the HDL pattern, so f(δH) becomes f(r) since there is spatial dependence of the HDL. Of these terms, Wsat plays the primary role in the overall linewidth, and the other terms determine the degree of roll-off of the line edges.
With the ultimate nanostructure fabrication, the ALD mask alone does not determine the total feature size. Instead, the pattern size depends upon the degree of erosion of the substrate under the mask. The total etched linewidth is expressed as Wt = Wm – We = Wsat + f(r) + M – We, where We indicates an erosion linewidth, or pattern size reduction due to the etching process. This depends upon, among other things, the thickness and quality of the etch mask as described above for Wm. For a case where the linewidth simply requires the removal of mask before etching occurs, the We term is zero, however it is observed that there is a modification to the feature size after etching relative to the mask shape, suggesting that more complicated dynamics are at play.
Of the elements determining linewidth limitations, Wsat can be reduced to a minimum width of ~4 nm before the growth stops appearing the same as bulk ALD.35 Of the other elements the mushroom growth effect, M (and as a consequence Wm), can only be reduced if the total film thickness is reduced, correlating with the total nanostructure height after etching. The line broadening effect due to the spatial dependence of the density of H remaining on the surface, f(δH), can be reduced to nearly zero by using multi-mode HDL which produces HDL line edges with negligible line edge roll-off.24 To demonstrate the effect of this reduction in f(δH), Figure 6 shows a pattern array of squares produced using multi-mode HDL. The array includes patterns with HDL linewidths of 7 nm, 14 nm, and 21 nm from top to bottom, and inner HDL opening sizes of 7 nm, 14 nm, and 21 nm from left to right. While there is a slight misalignment of the multi-mode HDL in the bottom row, along the top row the registration is precise to <1 nm. After RIE, the lines remain primarily intact to widths of 5 nm with two small defects, and the openings between lines are resolvable for all of the patterns with the 7 nm holes barely resolvable using this metrology tool.
Figure 6. Linewidth and holewidth test. (A) STM of HDL of boxes written using multi-mode HDL. The linewidth of rows is 21 nm, 14 nm, and 7 nm from the bottom to the top, respectively, and the holewidth of the columns is 7 nm, 14 nm, and 21 nm from left to right, respectively. (B) SEM of the same patterns after ALD and RIE. Please click here to view a larger version of this figure.
The ultimate limits of this process depend upon selectivity of the ALD process, the quality of the HDL, the resistance of the mask to etching, and the desired feature shapes themselves. Methods to improve selectivity based on chemistry and background defect mitigation has already been addressed above. It has been shown previously that leaving H defects in the patterned areas reduces the quality of mask growth, and thus the resistance to etching.35 Also, lack of careful control over the patterned line edges results in a mask “roll-off”, or excessive thinning of the mask along the edges of patterns which acts as a proximity effect preventing close placement of patterns. Fortunately, the selectivity of the etch process depends upon the mask thickness, so for spurious deposition on the background or defects along the edges of patterns the net effect is small. Furthermore, for structures shorter than 20 nm, thinner mask layers will likely be possible. Since the ALD growth occurs in a mushroom manner, thinner masks due to shorter structures will result in even better lateral control and smaller features than those demonstrated here. While the ultimate feature size reductions are not known for this process, certainly some down-scaling is likely.
While the SEM metrology leaves uncertainty regarding feature size and positioning, the first metrology step described in the top of Figure 1 gives atomic precision with regard to the HDL pattern as written. Since the Si(100)-H surface consists of a very regular lattice, and since the STM can be operated in a non-destructive imaging mode, the HDL patterns can be imaged without inducing further surface damage or further patterning, in contrast to other techniques such as e-beam lithography. With the atomic scale imaging of the invariant Si(100) lattice, the STM metrology eliminates the largest part of the positioning uncertainty related to the AFM and SEM metrology steps. In Figure 6B the box array appears skewed, for example. With the high resolution metrology of the STM giving atomic precision of the feature positions within the array, the apparent skew can be confirmed to be due to SEM imaging artifacts. Also, with highly precisely known spacings between the array features, an additional calibration uncertainty with respect to linewidths in the SEM images is eliminated.
This manuscript describes a nanofabrication method which utilizes the atomic precision of scanning tunneling microscope-based hydrogen depassivation lithography (HDL). HDL produces chemically reactive patterns on a Si(100)-H surface where atomic layer deposition of titania produces a localized etch mask with a lateral dimension demonstrated down to below 10 nm. Reactive ion etching then transfers the HDL patterns into the substrate, making 17 nm tall patterns with high precision lateral control. In order to achieve these results, samples must be protected during venting and transfer between instruments. With careful control of sample handling, nanostructures with traceability to the atomic lattice can be fabricated with atomic position precision and ~1 nm size precision.
The authors have nothing to disclose.
This work was supported by a Contract from DARPA (N66001-08-C-2040) and by a grant from the Emerging Technology Fund of the State of Texas. The authors would like to acknowledge Jiyoung Kim, Greg Mordi, Angela Azcatl, and Tom Scharf for their contributions related to selective atomic layer deposition, as well as Wallace Martin and Gordon Pollock for ex-situ sample processing.
Si Wafer | VA Semiconductor | P type (Boron) Si<100> +/- 2 degrees, 280 mm +/- 25 mm thick, 0.01-0.02 ohm-cm | |
Ta foil | Alfa Aesar | 335 | 0.025mm (0.001in) thick, 99.997% (metals basis) |
Methanol | Alfa Aesar | 19393 | Semiconductor Grade, 99.9% |
2-Propanol | Alfa Aesar | 19397 | Semiconductor Grade, 99.5% |
Acetone | Alfa Aesar | 19392 | Semiconductor Grade, 99.5% |
Argon | Praxair | Ultra high purity (grade 5.0) | |
Deionized water | Millipore | Milli-Q Water Purification System | >18 MW resistance water produced on demand. |
TiCl4 | Sigma Aldrigh | 254312 | ≥99.995% trace metals basis |
O2 | Matheson | G2182101 | Research Grade |
SF6 | Matheson | G2658922 | Ultra high purity (grade 4.7) |
Blue Medium Tack Roll | Semiconductor Equipment Corporation | 18074 | Thickness 75 um / .003” Length 200 M / 660’ |