- Email: [email protected]
Stronger silicon for microsystems
Available online at www.sciencedirect.com
Acta Materialia 58 (2010) 439–448 www.elsevier.com/locate/actamat
Stronger silicon for microsystems B.L. Boyce *, M.J. Shaw, P. Lu, M.T. Dugger Sandia National Laboratories, P.O. Box 5800, MS0889, Albuquerque, NM 87185-0889, USA Received 22 January 2009; received in revised form 4 September 2009; accepted 9 September 2009 Available online 8 October 2009
Abstract Few studies have deliberately varied the microstructure of microfabricated polycrystalline silicon (polySi) to examine their effects on resulting mechanical performance and reliability. In the present study, the tensile strength distributions of four microfabricated polySi variants were examined, corresponding to two different grain sizes (285 nm vs. 125 nm) in both the undoped and heavily P-doped conditions. Microtensile tests revealed that the coarse-grained materials exhibited significantly lower characteristic strengths (1.48–1.76 GPa) compared to the fine-grained material (2.80–2.83 GPa). The difference in strength was attributed largely to preferential etching of grain boundary grooves that were considerably more pronounced in the coarse-grained material. The presence of phosphorous doping had a less pronounced effect on strength values, lowering the characteristic strength of coarse-grained material by merely 16% and having little or no effect on the fine-grained material. Ó 2009 Acta Materialia Inc. Published by Elsevier Ltd. All rights reserved. Keywords: Lithography; Chemical vapor deposition; Fracture; Tension test; Ultrafine-grained microstructure
1. Introduction Much of the prior work on the development of MEMS processing routes has focused on fabricating structures with controlled dimensions and minimal residual stress. The processing techniques, derived largely from the microelectronics industry, have not been optimized with respect to structural properties such as modulus, fracture strength, fatigue resistance, and fracture toughness. The present study examines possible processing pathways to improve the tensile strength of polycrystalline silicon (polySi). The motivating hypothesis for this work is that the tensile strength of polySi can be improved by identifying processing conditions to minimize the extent of critical defects. Specifically, our initial expectation based on preliminary experiments and existing literature was that small changes in the microstructure and/or doping conditions could improve polySi’s mean strength by a factor of two or more by affecting the formation and size of failure–critical defects. Establishing connections between processing con*
Corresponding author. Tel.: +1 505 845 7525; fax: +1 505 844 7910. E-mail address: [email protected] (B.L. Boyce).
ditions, microstructure, critical defects, and structural reliability may assist in the future development of highly reliable MEMS materials. Prior work on the fracture strength of polySi has largely been limited to a few established processing methods such as the SUMMiT VTM process at Sandia National Labs and the PolyMUMPs process at MEMSCAP, Inc. For this reason, there have been relatively few studies which explicitly vary the microstructure (i.e. grain size, distribution and crystallographic texture) of microfabricated silicon to examine the effects of microstructure on properties such as fracture strength. An important exception is the work of Ballarini and co-workers [1], who compared the bending strength of amorphous silicon (a-Si) and polySi with a 100 nm grain size. They found that, on average, a-Si was nearly twice the strength of the polySi (9.7 GPa compared to 4.9 GPa average strength). This study highlights the notion that microstructure plays an important role in determining the resulting mechanical properties. Microstructure governs fracture strength via three possible avenues: (1) by modifying the material’s resistance to fracture, i.e. altering the toughening mechanisms;
1359-6454/$36.00 Ó 2009 Acta Materialia Inc. Published by Elsevier Ltd. All rights reserved. doi:10.1016/j.actamat.2009.09.022
440
B.L. Boyce et al. / Acta Materialia 58 (2010) 439–448
(2) by imparting local intragranular stress risers, most pronounced near grain boundaries and triple junctions, due to the constraining elastic crystallographic anisotropy of neighboring grains in the polycrystal; (3) by affecting the size, shape, distribution, or nature of failure-critical defects. 1.1. Resistance to fracture
Voronoi tessellated microstructure on the local strain energy release rate at the tip of a long crack. They found that materials with strong elastic anisotropy and non-cubic crystal symmetry could be sensitive to the local microstructure with energy release rates affected by up to 40% in extreme cases. However, for cubic materials with weak elastic anisotropy such as silicon, the elastic anisotropy of the microstructure was found to have a negligible effect on the local crack tip stresses or driving force for failure.
In Ballarini’s comparison of a-Si and polySi, fracture toughness was measured for both materials. In spite of their very different average strength values, the average fracture toughness of these two materials was p found to be statistically indistinguishable, 1.0 MPa m, suggesting that microstructure may not play an important role in toughness in this particular case. As a counter-example, Cho et al. [2] experimentally measured the fracture toughness values of several polysilicon tensile bars that had been fabricated by MEMSCAP’s Multi-User MEMS Processes (MUMPS) and precracked by radial corner cracks extending from nearby indentation in the sacrificial oxide prior to release. They found that the fracture toughness from specimen to p specimen varied by 50%, from 0.843 to 1.225 MPa m. This large variation in fracture toughness was attributed to local crack tip variations in crystallographic orientation, grain boundary toughness, and crack tip shielding conditions. Numerical cohesive zone modeling by Foulk et al. [3] of the grain-bridging effect in brittle microstructures also illustrates various scenarios where grain boundaries can influence the material’s macroscopic toughness. Under most typical silicon MEMS processes, doping is not expected to have a substantial effect on fracture toughness. Cook has recently shown, through indentation toughness experiments, that neither heavy n-type P-doping or p-type B-doping has a substantial effect on silicon’s intrinsic toughness [4]. Although doping levels can affect dislocation mobilities, there is little reason to believe that diffusionally driven dopant content would have a dramatic effect on the intrinsic toughness, unless the dopant poisons the grain boundaries causing a transition from transgranular to intergranular fracture. Son et al. [5] showed that the doping process can, under other conditions, have a notable effect on both fracture strength and fracture toughness. In comparing boron implantation, phosphorous implantation, and POCl3 diffusion, they found that the diffusion-doped material had the highest strength and fracture toughness. They suggest that the implantation method causes ion damage to the lattice, lower intrinsic toughness, and hence lower strength, which pervades even if the films are well annealed.
The critical strength-limiting defects in most brittle MEMS materials are surface flaws induced by either the chemical etching process [7] or annealing. In polySi, grain boundaries can be sites for critical defects [8]. Grain boundary grooves, also called grain boundary depressions [8], are the result of either localized preferential etching or thermal surface diffusion [9] and often represent the largest surface flaws. In single-crystal Si, Alan et al. [10] showed that the addition of a tetramethylammonium hydroxide (TMAH) etch after a standard potassium hydroxide (KOH) etch reduced the root-mean-squared surface roughness from 1.5 nm to 0.4 nm. This reduction in surface roughness increased the strength of the KOH + TMAH etched material by 25% over the baseline strength of the KOH-etched material. Similarly, Miller et al. [11] showed that the average strength of MEMSCAP SOIMUMPS single-crystal silicon could vary by 50%, depending on the extent of etch-induced sidewall and edge defects. The etch-induced surface defect structure and oxide thickness of both single-crystal and polycrystalline silicon has been shown to be strongly affected by anodic oxidation/galvanic corrosion when silicon is in intimate contact with a metal layer such as gold [12–15]. This galvanic corrosion can result in a highly porous, extensively oxidized surface layer, and the extent of surface damage has been shown to correlate well with observed degradation in the fracture strength, which can be reduced by >90% under extreme conditions [16,17]. Doping may also affect the size and shape of critical defects. Dopant content has long been known to affect etching or thermal grain-boundary grooving processes (see for example Ref. [18]). With few exceptions, the presence of dopant exacerbates the defect formation process, especially when the dopant is locally segregated to grain boundaries.
1.2. Local microstructural stresses
2.1. Polysilicon microfabrication
In a separate work, Ballarini et al. [6] simulated the role of polycrystalline elastic anisotropy associated with random crystallographic texture in an equiaxed Poisson–
The polysilicon materials for the present study were produced at Sandia National Laboratories’ Microfabrication Facility in Albuquerque, NM. Four wafers were examined
1.3. Critical defects
2. Method
B.L. Boyce et al. / Acta Materialia 58 (2010) 439–448
corresponding to two different grain sizes and two different doping levels1. The substrate and initial fabrication process was identical to the well-established SUMMiT VTM process [19]. Specifically, the starting handle wafers were single-crystal (100) n-type silicon with a 0.63 lm thermally grown oxide layer and a 0.8 lm low-pressure chemical vapor deposition (LPCVD) low-stress nitride (LSN) layer. A 0.3 lm doped polysilicon LPCVD layer (“Poly0”) was deposited as a base layer for the subsequent processing steps. On top of the Poly0 layer, a 2.0 lm layer (“SacOx1”) of either phosphosilicate glass (PSG, 2%) or undoped silane glass (USG) was deposited and lithographically patterned, depending on the desired doping level. The structural layer of interest (“Poly1”) was deposited onto the SacOx1 sacrificial glass layer and lithographically patterned. For the coarsegrained films, the Poly1 layer consisted of 1.0 lm of undoped polySi deposited at 580 °C in a LPCVD furnace. Alternatively, the fine-grained laminate Poly1 films were created by depositing alternate layers of 100 nm polySi at 610 °C, followed by 100 nm of amorphous silicon at 550 °C. A total 10 individual layers were deposited resulting in a laminate polySi film that was 1.0 lm thick. The Poly1 layers were covered with a second sacrificial oxide layer (“SacOx2”) of either 2% PSG or USG and chemomechanically planarized to 2.0 lm of additional thickness. The four wafers were all diffusion annealed at 1050 °C for 1 h in N2 and a second 1.0 lm layer of polySi (“Poly2”) was deposited at a temperature of 580 °C. The Poly2 layer merely served as a structural reinforcement in the ring-grip section of the tensile bars, and its properties were not examined in the present study. A 1.0 lm layer of either 2% PSG or USG was used as a hard mask for patterning the second polySi layer, the wafers were annealed at 1050 °C for 1 h in N2 to diffuse the phosphorous doping into the second polySi layer, and the second polySi layer was patterned. Finally, a 0.7 lm layer of Al–0.5 wt.% Cu was sputter-deposited on both wafers and patterned on bond pads and select structures. This metallic layer was not present on the tensile specimens nor was it in electrochemical contact with the test specimens, to avoid galvanic degradation of tensile properties. An example of the resulting layer stack is shown in the unreleased state in Fig. 1. The devices were released from their encapsulating oxide by etching in an HF-based chemistry with etch time adjusted to achieve the desired undercut followed by a supercritical CO2 drying process. 2.2. Tensile testing Tensile tests on microfabricated polysilicon structures were performed using the recently established “pull-tab” test method [20]. In this method, a ring on the free end of a
1 The four wafers presented in the current study are identified as Wafers 6, 8, 10 and 12 from reticle set RS636.
441
Fig. 1. TEM cross-section of the film stack, summarizing the layer-bylayer process flow. The Poly1 layer was customized (grain size, dopant level) and subsequently tested. This particular example corresponds to a Poly1 layer with coarse-grain size and 2.0% PSG content.
micromachined tensile bar is engaged with a tungsten probe tip that has been focused-ion-beam (FIB) machined into a right cylinder to fit inside the test specimen ring. The probe tip is servoelectrically actuated to pull the specimen in tension and force is measured with a Transducer Techniques 10 g load cell with a noise floor of 10 lN, corresponding to a stress resolution of 2.5 MPa. The displacement actuation rate of 0.5 lm s–1 corresponded to a stress rate of 230 MPa s–1. An example of the microfabricated tensile specimen design used in the present study is shown in Fig. 2. This specific design incorporated a symmetric pair of rings, both of which were affixed to the substrate via attached posts on either side of the rings. To liberate one end of the tensile specimen in this design, one ring’s attachment posts were intentionally broken prior to testing using a tungsten probe tip and micromanipulator. The “fixed” ring at the other end of the tensile bar was left attached to the substrate. Earlier versions of the “pull-tab” design for the SUMMiT VTM process incorporated a free-rotating hub design opposite to the ring end of the tensile bar to enable self-alignment [20]. The present custom-fabrication process was not amenable to a hub design, so the fixed end of the specimen was rigidly attached to the substrate. This approach was used in other recent studies and determined to have minimal impact on the resulting tensile properties if the cylindrical tungsten probe tip was precisely aligned and centered in the specimen ring [11,16,17]. Fracture strength was calculated as the maximum force at failure divided by the cross-sectional area of the tensile bar. Our prior work [20] originally relied on the intended
442
B.L. Boyce et al. / Acta Materialia 58 (2010) 439–448
Fig. 2. FESEM image of a typical double-ring tensile bar (45° tilt). The rings both consist of two stacked polysilicon layers for added rigidity whereas the gage section consists only of the lower polysilicon layer. As fabricated, the rings are attached to the substrate via posts on either side of each ring. Prior to testing, one ring is freed from the substrate by breaking the post attachments.
(as-drawn) dimensions of the tensile bars, rather than the actual as-fabricated dimensions. However, due to linewidth loss associated with the lithographic process, the true width of features fabricated in this process are typically 0.2 lm smaller than the intended nominal dimension. More recently, critical dimensions such as gage width and thickness were measured after fabrication and averaged to calculate the “true” fracture strength [11,16,17]. In the present study, the average dimensions of the tensile bars were measured in a field-emission scanning electron microscope (FESEM) separately for each wafer, and are included in Table 1. These wafer-delineated average dimensions were used in the reported calculations of fracture strength. 3. Results 3.1. Microstructure of custom-processed polysilicon Annular dark-field cross-sectional scanning transmission electron microscopy (STEM) images of the microstructure were obtained with a Tecnai-30 field-emission microscope operated at 300 keV. Dual-beam focused-ion beam (FIB) preparation of the TEM foils included a lowkeV cleaning step (5 keV or less) to minimize the incorporation of ion-induced damage to the TEM samples. Examples of the observed “coarse-grained” and “fine-grained laminate” microstructures are shown in Figs. 3 and 4, respectively. In both cases, the doping content did not have any discernable effect on the shape, size, or distribution of grains, as indicated quantitatively in Table 2. The coarsegrained Poly1 layer contained grains that occupied a significant fraction of the total thickness of the layer. The grains tended to be weakly columnar along the deposition direction. The fine-grained laminate Poly1 layer consisted of five distinct 200 nm sublayers. While the deposition conditions
Fig. 3. Two annular dark-field cross-sectional STEM images of a “coarsegrained” microstructure. Analysis of these and other images suggested that the typical grain size for the “coarse-grained” microstructures was 400 nm.
consisted of 10 layers (five 100 nm polycrystalline layers interspersed with five 100 nm amorphous layers), each of the amorphous layers had crystallized coherently with the neighboring crystalline layer, leaving little or no trace of the five amorphous interlayers. The five-layer laminate structure resulted in an approximately equiaxed grain structure with a distinctly smaller grain size than the coarse-grained counterparts. Average grain sizes were estimated for each of the four conditions based on STEM images. The complex transmission images rendered traditional stereology techniques such as the chord intercept method impractical. Instead, for each condition, the major and minor diameters of 20 clearly delineated grains were measured. For each grain, the “average” diameter was estimated as the average of the major and minor diameters. The resulting grain size measures are summarized in Table 2. 3.2. Observed strength distributions For each of the four polysilicon conditions studied, at least 20 tensile specimens were tested to establish a minimally biased estimate of the true Weibull modulus [21]. The observed strength values were rank ordered from
Table 1 As-fabricated tensile bar cross-sectional dimensions (average ± 1 standard deviation). Wafer
Coarse-grained 2.0% PSG
Fine-grained 2.0% PSG
Coarse-grained undoped
Fine-grained undoped
Thickness (nm) Width (nm)
1029 ± 23 3745 ± 30
999 ± 18 3819 ± 26
958 ± 29 3854 ± 17
970 ± 15 3922 ± 61
B.L. Boyce et al. / Acta Materialia 58 (2010) 439–448
Fig. 4. Two annular dark-field cross-sectional STEM images of a “finegrained laminate” microstructure. Analysis of these and other images suggested that the typical grain size for the “fine-grained” microstructures was 150 nm.
n = 1, . . ., N where N was the total number of samples for that specific condition. The probability estimator P used for this analysis was P¼
n 0:5 N
while this estimator tends to slightly overestimate the Weibull modulus, it has been shown to give the least biased estimate of the modulus for sample sizes N P 20 [21,22]. The observed strength distributions for the four polysilicon conditions are shown in Fig. 5. The characteristic strength rh and Weibull modulus m for each of the four conditions was estimated in Table 3 using a two-parameter Weibull distribution for the probability of failure P as a function of applied stress r. m r P ¼ 1 exp rh 3.3. Surface topography Scanning electron microscopy revealed very distinct surface topography for the coarse-grained and fine-grained
443
Fig. 5. Observed strength distributions for each of the four Poly1 variants. Tests were conducted in air at room temperature. The error bars are not visible since the stress resolution (±2.5 MPa) was far smaller than the width of the data points (50 MPa).
variants, as shown in Fig. 6. The coarse-grained wafers exhibited well-defined grain boundary grooves that encircle some but not all of the grains on the top-surface. Also present are small (<50 nm) nodules or bumps. These bumps were previously thought to be SiC particles that result from migration and precipitation of residual carbon from the tetraethylorthosilicate (TEOS) sacrificial oxide during thermal annealing [23]. However, in the present study either undoped silane glass or 2% PSG was used in lieu of TEOS and there was no carbon source for the formation of SiC particles. We speculate that the nodules or bumps may instead be the nucleus of new grains initiating on the surface. The top-surface grain boundary grooves appear to be the deepest and sharpest notch-like features on the surfaces of the coarse-grained material and thus are expected to be the critical flaws that drive failure. In the case of the finegrained laminate structures, the top-surface is considerably smoother with no obvious preferential grain-boundary etching or nodular features. The much finer top-surface is not simply due to the smaller grain size: the average grain diameter of the fine-grained laminate is 125 nm whereas the top-surface features in the fine-grained laminate structures are much finer than 125 nm.
Table 2 Grain size measures (average ± 1 standard deviation). Wafer
Coarse-graine 2.0% PSG
Fine-grained 2.0% PSG
Coarse-grained undoped
Fine-grained undoped
Major diameter (nm) Minor diameter (nm) Avg. diameter (nm)
408 ± 227 160 ± 62 284 ± 137
153 ± 46 96 ± 24 125 ± 33
398 ± 184 181 ± 63 290 ± 115
158 ± 48 97 ± 38 128 ± 37
444
B.L. Boyce et al. / Acta Materialia 58 (2010) 439–448
Table 3 Weibull fit parameters for the four conditions. Wafer
Coarse-grained 2.0% PSG
Fine-grained 2.0% PSG
Coarse-grained undoped
Fine-grained undoped
Characteristic Strength, rh (GPa) Weibull modulus, m
1.48 8.7
2.83 13.5
1.76 12.9
2.80 8.6
Fig. 6. Oblique (45°) SEM view of the sidewall and top-surface of the four Poly1 variants. In Wafers 6 and 10, some top-surface grain boundaries are encircled by grain boundary grooves caused by preferential etching. Sidewall or corner notches indicated by arrows in were seen consistently in the heavily doped coarse-grained material and were not as extensive in the undoped coarse-grained material.
The differences in top-surface morphology and roughness were examined quantitatively by contact-mode atomic force microscopy (AFM) using a Park Scientific AutoProbe LS AFM with a 1–3 nm radius silicon stylus. For each of the four polySi variants, three to four 10 lm 10 lm regions were surveyed with a 256 256 pixel grid. Within each region, smaller 5 lm 5 lm and 2 lm 2 lm subregions were surveyed for increased resolution. Typical AFM images from the 5 lm scans are shown in Fig. 7. The distinction between the coarse-grained and fine-grained laminate variants is clearly consistent with the SEM images of Fig. 6. The resulting root-mean-squared (rms) surface roughness metric for each variant is shown in Table 4. The fine-grained laminate variants had 20% lower RMS roughness than the coarse-grained variants, consistent with the observation that the fine-grained laminates were stronger than the coarse-grained variants. The effect of phosphorous doping on the resulting RMS roughness was less significant and possibly even negligible compared to the effect of the grain size/structure. However, fracture strength is not directly related to average roughness over a field, but rather to the worst-case flaws in the material. An attempt was made to identify worst-case flaws in the surveyed regions, by obtaining high resolution images of the deepest features in the 10 lm regions surveyed. However, the observed “deepest” trenches, which were 20– 50 nm in depth, did not show any notable correlation with the strengths of the four variants. This apparent lack of correlation may be due to insufficient sampling. While AFM examination of the top-surface did not reveal any clear evidence of a doping-dependent change in flaw morphology, Fig. 6 shows that the coarse-grained 2.0% PSG
variant contained significant corner and sidewall defects (arrows in Fig. 6) that were not observed in the coarsegrained undoped variant. This qualitative observation is consistent with the mechanical strength in that undoped coarse-grained material was stronger than the doped coarse-grained material. In both the coarse-grained and fine-grained laminate materials, the sidewall edge appeared to be smoother than the top-surface. The sidewalls of the coarse-grained material exhibited vertical waviness or “curtaining”. In the heavily doped coarse-grained wafer, the curtains resulted occasionally in visible notch-like features at the top corner where grain boundary grooves intersected the sidewall (arrows in Fig. 6). The sidewalls of the fine-grained laminate material were notably smoother with less significant curtaining and no visible flaws compared to the coarsegrained material. 3.4. Fractography Typical fracture surfaces for the fine-grained laminate and coarse-grained materials are shown in Fig. 8. The observed fractography was consistent with the previous descriptions of microstructure and surface flaw topography. The coarse-grained material exhibited coarse crystallographic faceting corresponding to the larger grain size (i.e. only a few grains through-thickness). By comparison, the fracture surface of the fine-grained laminate exhibited much finer surface features with some horizontal delineation of the five-layer lamellar microstructure visible in the fracture surfaces. The fracture surface in Fig. 8b was tilted at an angle away from the expected mode-I plane, suggesting
B.L. Boyce et al. / Acta Materialia 58 (2010) 439–448
445
Fig. 7. Surface morphology of the four Poly1 variants as measured by atomic force microscopy.
Table 4 Surface roughness as measured by AFM. Wafer
6 coarse-grained 2.0% PSG
8 fine-grained 2.0% PSG
10 coarse-grained undoped
12 fine-grained undoped
5 lm 5 lm regions Surveyed RMS roughness (nm)
3 8.06 ± 0.11
4 6.92 ± 0.31
3 8.26 ± 0.44
4 6.22 ± 0.15
the possibility of some bending forces superimposed on the tensile forces. Crack initiation in brittle materials is often surmised from either obvious defects found on the fracture surface or from tracing back characteristic mirror–mist–hackle markings to the origin. In these micro-scale fracture surfaces, identifying failure origins is particularly challenging because: (a) traditional mirror–mist–hackle markings often do not develop over such a small scale, (b) the expected nano-scale defects (10–100 nm) are difficult to image by electron optics in a poorly conductive medium and (c) the crystallographic faceting is of a similar scale to the features of interest, thereby convoluting interpretation. Nevertheless, in the coarse-grained fractograph shown in Fig. 8a, the critical defect is likely the deep trench near the right sidewall of the specimen. The trench appears to be caused by excessive preferential etching of the grain boundary. In
the fine-grained laminate structure, no obvious defect was identified in the fracture surface. 4. Discussion 4.1. The role of grain boundary grooves The critical strength-limiting defects in the coarse-grained material appear to be grooves caused by preferential etching of grain boundaries. These grain boundary grooves are typical for common polySi processes. Miller et al. [16] has shown that the grain boundaries of MUMPs polySi appear to be a preferred path for structural degradation. When galvanic corrosion is induced by etching a multi-layer structure that is in electrical contact with gold, preferential grainboundary attack appears to be responsible for a transition from transgranular fracture to intergranular fracture.
446
B.L. Boyce et al. / Acta Materialia 58 (2010) 439–448
process shown in Fig. 10 confirm the direct correspondence between observed surface grooves and underlying grain boundaries. However, not all grain boundaries are attacked equally. Clearly, in the coarse-grained custom material of the present study, the grooves appear to outline only certain grains. Crystallographic texture analysis is ongoing to investigate the possible correlation between the deepest grooves and grain boundaries with the highest surface energy. While grain boundary grooves appear to be the sharpest, deepest notch-like feature in the coarse-grained material, it is not clear that they are responsible for failure in the fine-grained material. Nevertheless, the absence of well-delineated grain boundary grooves in the fine-grained laminate appears to enhance the resulting tensile properties: the characteristic fracture strength for the fine-grained laminate was 60–90% higher than that for the coarse-grained material. 4.2. The role of phosphorous doping
Fig. 8. Typical fracture surfaces for coarse-grained and fine-grained laminate materials.
SUMMiT polysilicon also shows evidence of preferential grain-boundary attack, as is evident from the grooves shown in Fig. 9. Cross-sectional TEM images from the SUMMiT VTM
The addition of phosphorous (n-type) doping lowered the average fracture strength of the coarse-grained polysilicon by 16%. This is a similar result to that of Ballarini [1] where boron (p-type) doping was found to lower the fracture strength by 12% compared to undoped material. Interestingly, the phosphorous doping had little or no effect on the fine-grained laminate material. The important distinction here is that the coarse-grained material showed clear grain boundary grooves at the top-surface whereas the smoother topography of the fine-grained material did not clearly correlate with grain boundaries. This suggests that the diffusion-driven dopant content in the present study does not likely alter the intrinsic toughness of the material [4], which would have affected both microstructures equally, but more likely, the dopant may exacerbate the localized grooving process. It is clear from Fig. 6 that the heavily doped coarse-grain material exhibited more pronounced crevices at the corners and sidewall surfaces than the undoped counterpart. 4.3. Statistical variability in fracture strength: implications for reliability
Fig. 9. In this fractured SUMMiT V polysilicon tensile bar, grain boundary grooves are visible both on the top-surface and on the sidewalls.
The Weibull modulus, determined for each of the four silicon variants in Table 3, describes the breadth of the statistical distribution in failure strengths: low values of Weibull modulus correspond to high degrees of scatter. Not only does the doped fine-grained laminate material have the highest characteristic strength of 2.83 GPa, but it also has the highest Weibull modulus value corresponding to the tightest distribution in fracture strengths. This has important implications for yield and reliability. For example, if an acceptable probability of failure was 1%, the doped fine-grained laminate material with a Weibull modulus of 13.5 would have an allowable design stress of 2.01 GPa or 71% of its 2.83 GPa characteristic strength. By comparison, the doped coarse-grained material with a Weibull modulus of 8.7 would have an allowable stress of
B.L. Boyce et al. / Acta Materialia 58 (2010) 439–448
447
Fig. 10. (left image) Annular dark-field TEM image of a sidewall surface crevice penetrating at a grain boundary and (right image) higher resolution bright-field TEM image of another sidewall crevice penetrating at a grain boundary. Some grain boundaries showed no signs of preferential attack.
0.872 GPa or 59% of its 1.48 GPa characteristic strength. For reliable design, the statistical scatter in fracture strengths can be as important as the average or characteristic strength of the material. The statistical scatter is thought to be related to the homogeneity of the flaw population, and possibly the homogeneity in the intrinsic fracture toughness. A more detailed discussion of this issue and its implications is presented in Ref. [24]. 4.4. Other considerations: role of environment Silicon fracture also involves an important environmental aspect that is not clearly understood. For example, Alan et al. [25] has shown that the strength of single-crystal Si with hydrogen-terminated surfaces degrades by as much as 30% during the first month of air exposure after release. However, microfabricated single-crystal Si that has been coated with a methyl monolayer is initially stronger than H-terminated single-crystal Si and shows no degradation after a month of exposure to air. These observations may be attributed to stress–corrosion cracking that can occur in stressed surface oxide films [26], which may form easier on H-terminated surfaces than on monolayer-protected surfaces. However, this mechanism is somewhat speculative and does not clearly identify the precise mechanism by which the strength is altered (i.e. by altering the defect morphology or the local resistance to fracture). 4.5. Optimizing the mechanical performance of silicon MEMS The present study illustrates the significant role that processing has on the resulting strength of polycrystalline silicon MEMS. By merely switching from a single-layer deposition process to a multi-layer laminate process, the strength of silicon was increased by a factor of two. However, we have by no means optimized the strength of silicon. Further, while we see that the processing conditions can affect the location, shape, and size of failure-critical
defects, we still have only very limited understanding of the processing–structure relationships that govern the formation of critical defects. While the present study has focused solely on the tensile strength of polySi, there are many other mechanical properties that can be relevant to MEMS components, including Young’s modulus, fracture toughness, fatigue resistance, wear resistance, etc. There is a continuing need for improved understanding of the role that processing plays on the microstructure and resulting properties of MEMS materials. As MEMS technology continues to evolve from its microelectronics origins, structural optimization studies for enhanced mechanical performance will enable a broader range of highreliability MEMS applications. 5. Summary and conclusions In this study, we observed the fracture strength distributions of phosphorous diffusion-doped and undoped polycrystalline silicon with either a 285-nm grained equiaxed microstructure or a 125-nm grained laminate microstructure. The coarse-grained materials exhibited significantly lower characteristic strengths (1.48–1.76 GPa) compared to the fine-grained laminate material (2.80–2.83 GPa). The difference in strength was attributed largely to preferential etching of grain boundary grooves that were considerably more pronounced in the coarse-grained material. The lack of significant grain-boundary grooving in the fine-grained laminate material may be related to the alternating deposition process which incorporated an “amorphous” deposition step. The presence of phosphorous doping lowered the characteristic strength of coarsegrained material by merely 16% and had little or no effect on the fine-grained material. Acknowledgements The authors would like to thank Dr L. Phinney for collaborating on the fabrication of these custom polycrystal-
448
B.L. Boyce et al. / Acta Materialia 58 (2010) 439–448
line silicon materials. The assistance of T. Crenshaw and B. McKenzie is greatly appreciated in support of mechanical testing and SEM imaging, respectively. BLB would like to thank Drs E.D. Reedy Jr., J. Foulk, M.P De Boer, T.E. Buchheit and J.S. Custer for useful discussions on this topic. Sandia is a multiprogram laboratory operated by Sandia Corporation, a Lockheed Martin Company, for the United States Department of Energy’s National Nuclear Security Administration under contract DE-AC0494AL85000. References [1] Ballarini R, Kahn H, Tayebi N, Heuer AH. Effects of microstructure on the strength and fracture toughness of polysilicon: a wafer level testing approach. In: Muhlstein CLBSB, editor. Symposium on mechanical properties of structural films. Orlando, (FL): American Society Testing and Materials; 2000. p. 37. [2] Cho SW, Jonnalagadda K, Chasiotis I. Fatigue Fract Eng Mater Struct 2007;30:21. [3] Foulk JW, Johnson GC, Klein PA, Ritchie RO. J Mech Phys Solids 2008;56:2381. [4] Cook RF. J Mater Sci 2006;41:841. [5] Son D, Kim JJ, Lim TW, Kwon D. Thin Solid Films 2004;468:167. [6] Ballarini R, Mullen RL, Heuer AH. Int J Fract 1999;95:19.
[7] Alsem DH, Boyce BL, Stach EA, Ritchie RO. Sens Actuator A— Phys 2008;147:553. [8] Chasiotis I, Knauss WG. J Mech Phys Solids 2003;51:1533. [9] Robertson WM. J Am Ceram Soc 1981;64:9. [10] Alan T, Hines MA, Zehnder AT. Appl Phys Lett 2006:89. [11] Miller DC, Boyce BL, Dugger MT, Buchheit TE, Gall K. Sensors Actuators: A Phys 2007;138:130. [12] Kahn H, Deeb C, Chasiotis I, Heuer AH. J Microelectromech Syst 2005;14:914. [13] Pierron ON, Macdonald DD, Muhlstein CL. Appl Phys Lett 2005:86. [14] Miller DC, Becker CR, Stoldt CR. J Electrochem Soc 2008;155:F253. [15] Miller DC, Gall K, Stoldt CR. Electrochem Solid State Lett 2005;8:G223. [16] Miller DC, Boyce BL, Gall K, Stoldt CR. Appl Phys Lett 2007:90. [17] Miller DC, Boyce BL, Kotula PG, Stoldt CR. J Appl Phys 2008:103. [18] McAllister PV, Cutler IB. J Am Ceram Soc 1972;55:351. [19] Sniegowski JJ, de Boer MP. Annu Rev Mater Sci 2000;30:299. [20] Boyce BL, Grazier JM, Buchheit TE, Shaw MJ. J Microelectromech Syst 2007;16:179. [21] Sullivan JD, Lauzon PH. J Mater Sci Lett 1986;5:1245. [22] Bergman B. J Mater Sci Lett 1984;3:689. [23] DelRio FW, Dunn ML, Boyce BL, Corwin AD, de Boer MP. T J Appl Phys 2006:99. [24] Boyce BL, Ballarini R, Chasiotis I. J Micromech Microeng 2008:18. [25] Alan T, Zehnder AT, Sengupta D, Hines MA. Appl Phys Lett 2006:89. [26] Kahn H, Avishai A, Ballarini R, Heuer AH. Scripta Mater 2008;59:912.
Acta Materialia 58 (2010) 439–448 www.elsevier.com/locate/actamat
Stronger silicon for microsystems B.L. Boyce *, M.J. Shaw, P. Lu, M.T. Dugger Sandia National Laboratories, P.O. Box 5800, MS0889, Albuquerque, NM 87185-0889, USA Received 22 January 2009; received in revised form 4 September 2009; accepted 9 September 2009 Available online 8 October 2009
Abstract Few studies have deliberately varied the microstructure of microfabricated polycrystalline silicon (polySi) to examine their effects on resulting mechanical performance and reliability. In the present study, the tensile strength distributions of four microfabricated polySi variants were examined, corresponding to two different grain sizes (285 nm vs. 125 nm) in both the undoped and heavily P-doped conditions. Microtensile tests revealed that the coarse-grained materials exhibited significantly lower characteristic strengths (1.48–1.76 GPa) compared to the fine-grained material (2.80–2.83 GPa). The difference in strength was attributed largely to preferential etching of grain boundary grooves that were considerably more pronounced in the coarse-grained material. The presence of phosphorous doping had a less pronounced effect on strength values, lowering the characteristic strength of coarse-grained material by merely 16% and having little or no effect on the fine-grained material. Ó 2009 Acta Materialia Inc. Published by Elsevier Ltd. All rights reserved. Keywords: Lithography; Chemical vapor deposition; Fracture; Tension test; Ultrafine-grained microstructure
1. Introduction Much of the prior work on the development of MEMS processing routes has focused on fabricating structures with controlled dimensions and minimal residual stress. The processing techniques, derived largely from the microelectronics industry, have not been optimized with respect to structural properties such as modulus, fracture strength, fatigue resistance, and fracture toughness. The present study examines possible processing pathways to improve the tensile strength of polycrystalline silicon (polySi). The motivating hypothesis for this work is that the tensile strength of polySi can be improved by identifying processing conditions to minimize the extent of critical defects. Specifically, our initial expectation based on preliminary experiments and existing literature was that small changes in the microstructure and/or doping conditions could improve polySi’s mean strength by a factor of two or more by affecting the formation and size of failure–critical defects. Establishing connections between processing con*
Corresponding author. Tel.: +1 505 845 7525; fax: +1 505 844 7910. E-mail address: [email protected] (B.L. Boyce).
ditions, microstructure, critical defects, and structural reliability may assist in the future development of highly reliable MEMS materials. Prior work on the fracture strength of polySi has largely been limited to a few established processing methods such as the SUMMiT VTM process at Sandia National Labs and the PolyMUMPs process at MEMSCAP, Inc. For this reason, there have been relatively few studies which explicitly vary the microstructure (i.e. grain size, distribution and crystallographic texture) of microfabricated silicon to examine the effects of microstructure on properties such as fracture strength. An important exception is the work of Ballarini and co-workers [1], who compared the bending strength of amorphous silicon (a-Si) and polySi with a 100 nm grain size. They found that, on average, a-Si was nearly twice the strength of the polySi (9.7 GPa compared to 4.9 GPa average strength). This study highlights the notion that microstructure plays an important role in determining the resulting mechanical properties. Microstructure governs fracture strength via three possible avenues: (1) by modifying the material’s resistance to fracture, i.e. altering the toughening mechanisms;
1359-6454/$36.00 Ó 2009 Acta Materialia Inc. Published by Elsevier Ltd. All rights reserved. doi:10.1016/j.actamat.2009.09.022
440
B.L. Boyce et al. / Acta Materialia 58 (2010) 439–448
(2) by imparting local intragranular stress risers, most pronounced near grain boundaries and triple junctions, due to the constraining elastic crystallographic anisotropy of neighboring grains in the polycrystal; (3) by affecting the size, shape, distribution, or nature of failure-critical defects. 1.1. Resistance to fracture
Voronoi tessellated microstructure on the local strain energy release rate at the tip of a long crack. They found that materials with strong elastic anisotropy and non-cubic crystal symmetry could be sensitive to the local microstructure with energy release rates affected by up to 40% in extreme cases. However, for cubic materials with weak elastic anisotropy such as silicon, the elastic anisotropy of the microstructure was found to have a negligible effect on the local crack tip stresses or driving force for failure.
In Ballarini’s comparison of a-Si and polySi, fracture toughness was measured for both materials. In spite of their very different average strength values, the average fracture toughness of these two materials was p found to be statistically indistinguishable, 1.0 MPa m, suggesting that microstructure may not play an important role in toughness in this particular case. As a counter-example, Cho et al. [2] experimentally measured the fracture toughness values of several polysilicon tensile bars that had been fabricated by MEMSCAP’s Multi-User MEMS Processes (MUMPS) and precracked by radial corner cracks extending from nearby indentation in the sacrificial oxide prior to release. They found that the fracture toughness from specimen to p specimen varied by 50%, from 0.843 to 1.225 MPa m. This large variation in fracture toughness was attributed to local crack tip variations in crystallographic orientation, grain boundary toughness, and crack tip shielding conditions. Numerical cohesive zone modeling by Foulk et al. [3] of the grain-bridging effect in brittle microstructures also illustrates various scenarios where grain boundaries can influence the material’s macroscopic toughness. Under most typical silicon MEMS processes, doping is not expected to have a substantial effect on fracture toughness. Cook has recently shown, through indentation toughness experiments, that neither heavy n-type P-doping or p-type B-doping has a substantial effect on silicon’s intrinsic toughness [4]. Although doping levels can affect dislocation mobilities, there is little reason to believe that diffusionally driven dopant content would have a dramatic effect on the intrinsic toughness, unless the dopant poisons the grain boundaries causing a transition from transgranular to intergranular fracture. Son et al. [5] showed that the doping process can, under other conditions, have a notable effect on both fracture strength and fracture toughness. In comparing boron implantation, phosphorous implantation, and POCl3 diffusion, they found that the diffusion-doped material had the highest strength and fracture toughness. They suggest that the implantation method causes ion damage to the lattice, lower intrinsic toughness, and hence lower strength, which pervades even if the films are well annealed.
The critical strength-limiting defects in most brittle MEMS materials are surface flaws induced by either the chemical etching process [7] or annealing. In polySi, grain boundaries can be sites for critical defects [8]. Grain boundary grooves, also called grain boundary depressions [8], are the result of either localized preferential etching or thermal surface diffusion [9] and often represent the largest surface flaws. In single-crystal Si, Alan et al. [10] showed that the addition of a tetramethylammonium hydroxide (TMAH) etch after a standard potassium hydroxide (KOH) etch reduced the root-mean-squared surface roughness from 1.5 nm to 0.4 nm. This reduction in surface roughness increased the strength of the KOH + TMAH etched material by 25% over the baseline strength of the KOH-etched material. Similarly, Miller et al. [11] showed that the average strength of MEMSCAP SOIMUMPS single-crystal silicon could vary by 50%, depending on the extent of etch-induced sidewall and edge defects. The etch-induced surface defect structure and oxide thickness of both single-crystal and polycrystalline silicon has been shown to be strongly affected by anodic oxidation/galvanic corrosion when silicon is in intimate contact with a metal layer such as gold [12–15]. This galvanic corrosion can result in a highly porous, extensively oxidized surface layer, and the extent of surface damage has been shown to correlate well with observed degradation in the fracture strength, which can be reduced by >90% under extreme conditions [16,17]. Doping may also affect the size and shape of critical defects. Dopant content has long been known to affect etching or thermal grain-boundary grooving processes (see for example Ref. [18]). With few exceptions, the presence of dopant exacerbates the defect formation process, especially when the dopant is locally segregated to grain boundaries.
1.2. Local microstructural stresses
2.1. Polysilicon microfabrication
In a separate work, Ballarini et al. [6] simulated the role of polycrystalline elastic anisotropy associated with random crystallographic texture in an equiaxed Poisson–
The polysilicon materials for the present study were produced at Sandia National Laboratories’ Microfabrication Facility in Albuquerque, NM. Four wafers were examined
1.3. Critical defects
2. Method
B.L. Boyce et al. / Acta Materialia 58 (2010) 439–448
corresponding to two different grain sizes and two different doping levels1. The substrate and initial fabrication process was identical to the well-established SUMMiT VTM process [19]. Specifically, the starting handle wafers were single-crystal (100) n-type silicon with a 0.63 lm thermally grown oxide layer and a 0.8 lm low-pressure chemical vapor deposition (LPCVD) low-stress nitride (LSN) layer. A 0.3 lm doped polysilicon LPCVD layer (“Poly0”) was deposited as a base layer for the subsequent processing steps. On top of the Poly0 layer, a 2.0 lm layer (“SacOx1”) of either phosphosilicate glass (PSG, 2%) or undoped silane glass (USG) was deposited and lithographically patterned, depending on the desired doping level. The structural layer of interest (“Poly1”) was deposited onto the SacOx1 sacrificial glass layer and lithographically patterned. For the coarsegrained films, the Poly1 layer consisted of 1.0 lm of undoped polySi deposited at 580 °C in a LPCVD furnace. Alternatively, the fine-grained laminate Poly1 films were created by depositing alternate layers of 100 nm polySi at 610 °C, followed by 100 nm of amorphous silicon at 550 °C. A total 10 individual layers were deposited resulting in a laminate polySi film that was 1.0 lm thick. The Poly1 layers were covered with a second sacrificial oxide layer (“SacOx2”) of either 2% PSG or USG and chemomechanically planarized to 2.0 lm of additional thickness. The four wafers were all diffusion annealed at 1050 °C for 1 h in N2 and a second 1.0 lm layer of polySi (“Poly2”) was deposited at a temperature of 580 °C. The Poly2 layer merely served as a structural reinforcement in the ring-grip section of the tensile bars, and its properties were not examined in the present study. A 1.0 lm layer of either 2% PSG or USG was used as a hard mask for patterning the second polySi layer, the wafers were annealed at 1050 °C for 1 h in N2 to diffuse the phosphorous doping into the second polySi layer, and the second polySi layer was patterned. Finally, a 0.7 lm layer of Al–0.5 wt.% Cu was sputter-deposited on both wafers and patterned on bond pads and select structures. This metallic layer was not present on the tensile specimens nor was it in electrochemical contact with the test specimens, to avoid galvanic degradation of tensile properties. An example of the resulting layer stack is shown in the unreleased state in Fig. 1. The devices were released from their encapsulating oxide by etching in an HF-based chemistry with etch time adjusted to achieve the desired undercut followed by a supercritical CO2 drying process. 2.2. Tensile testing Tensile tests on microfabricated polysilicon structures were performed using the recently established “pull-tab” test method [20]. In this method, a ring on the free end of a
1 The four wafers presented in the current study are identified as Wafers 6, 8, 10 and 12 from reticle set RS636.
441
Fig. 1. TEM cross-section of the film stack, summarizing the layer-bylayer process flow. The Poly1 layer was customized (grain size, dopant level) and subsequently tested. This particular example corresponds to a Poly1 layer with coarse-grain size and 2.0% PSG content.
micromachined tensile bar is engaged with a tungsten probe tip that has been focused-ion-beam (FIB) machined into a right cylinder to fit inside the test specimen ring. The probe tip is servoelectrically actuated to pull the specimen in tension and force is measured with a Transducer Techniques 10 g load cell with a noise floor of 10 lN, corresponding to a stress resolution of 2.5 MPa. The displacement actuation rate of 0.5 lm s–1 corresponded to a stress rate of 230 MPa s–1. An example of the microfabricated tensile specimen design used in the present study is shown in Fig. 2. This specific design incorporated a symmetric pair of rings, both of which were affixed to the substrate via attached posts on either side of the rings. To liberate one end of the tensile specimen in this design, one ring’s attachment posts were intentionally broken prior to testing using a tungsten probe tip and micromanipulator. The “fixed” ring at the other end of the tensile bar was left attached to the substrate. Earlier versions of the “pull-tab” design for the SUMMiT VTM process incorporated a free-rotating hub design opposite to the ring end of the tensile bar to enable self-alignment [20]. The present custom-fabrication process was not amenable to a hub design, so the fixed end of the specimen was rigidly attached to the substrate. This approach was used in other recent studies and determined to have minimal impact on the resulting tensile properties if the cylindrical tungsten probe tip was precisely aligned and centered in the specimen ring [11,16,17]. Fracture strength was calculated as the maximum force at failure divided by the cross-sectional area of the tensile bar. Our prior work [20] originally relied on the intended
442
B.L. Boyce et al. / Acta Materialia 58 (2010) 439–448
Fig. 2. FESEM image of a typical double-ring tensile bar (45° tilt). The rings both consist of two stacked polysilicon layers for added rigidity whereas the gage section consists only of the lower polysilicon layer. As fabricated, the rings are attached to the substrate via posts on either side of each ring. Prior to testing, one ring is freed from the substrate by breaking the post attachments.
(as-drawn) dimensions of the tensile bars, rather than the actual as-fabricated dimensions. However, due to linewidth loss associated with the lithographic process, the true width of features fabricated in this process are typically 0.2 lm smaller than the intended nominal dimension. More recently, critical dimensions such as gage width and thickness were measured after fabrication and averaged to calculate the “true” fracture strength [11,16,17]. In the present study, the average dimensions of the tensile bars were measured in a field-emission scanning electron microscope (FESEM) separately for each wafer, and are included in Table 1. These wafer-delineated average dimensions were used in the reported calculations of fracture strength. 3. Results 3.1. Microstructure of custom-processed polysilicon Annular dark-field cross-sectional scanning transmission electron microscopy (STEM) images of the microstructure were obtained with a Tecnai-30 field-emission microscope operated at 300 keV. Dual-beam focused-ion beam (FIB) preparation of the TEM foils included a lowkeV cleaning step (5 keV or less) to minimize the incorporation of ion-induced damage to the TEM samples. Examples of the observed “coarse-grained” and “fine-grained laminate” microstructures are shown in Figs. 3 and 4, respectively. In both cases, the doping content did not have any discernable effect on the shape, size, or distribution of grains, as indicated quantitatively in Table 2. The coarsegrained Poly1 layer contained grains that occupied a significant fraction of the total thickness of the layer. The grains tended to be weakly columnar along the deposition direction. The fine-grained laminate Poly1 layer consisted of five distinct 200 nm sublayers. While the deposition conditions
Fig. 3. Two annular dark-field cross-sectional STEM images of a “coarsegrained” microstructure. Analysis of these and other images suggested that the typical grain size for the “coarse-grained” microstructures was 400 nm.
consisted of 10 layers (five 100 nm polycrystalline layers interspersed with five 100 nm amorphous layers), each of the amorphous layers had crystallized coherently with the neighboring crystalline layer, leaving little or no trace of the five amorphous interlayers. The five-layer laminate structure resulted in an approximately equiaxed grain structure with a distinctly smaller grain size than the coarse-grained counterparts. Average grain sizes were estimated for each of the four conditions based on STEM images. The complex transmission images rendered traditional stereology techniques such as the chord intercept method impractical. Instead, for each condition, the major and minor diameters of 20 clearly delineated grains were measured. For each grain, the “average” diameter was estimated as the average of the major and minor diameters. The resulting grain size measures are summarized in Table 2. 3.2. Observed strength distributions For each of the four polysilicon conditions studied, at least 20 tensile specimens were tested to establish a minimally biased estimate of the true Weibull modulus [21]. The observed strength values were rank ordered from
Table 1 As-fabricated tensile bar cross-sectional dimensions (average ± 1 standard deviation). Wafer
Coarse-grained 2.0% PSG
Fine-grained 2.0% PSG
Coarse-grained undoped
Fine-grained undoped
Thickness (nm) Width (nm)
1029 ± 23 3745 ± 30
999 ± 18 3819 ± 26
958 ± 29 3854 ± 17
970 ± 15 3922 ± 61
B.L. Boyce et al. / Acta Materialia 58 (2010) 439–448
Fig. 4. Two annular dark-field cross-sectional STEM images of a “finegrained laminate” microstructure. Analysis of these and other images suggested that the typical grain size for the “fine-grained” microstructures was 150 nm.
n = 1, . . ., N where N was the total number of samples for that specific condition. The probability estimator P used for this analysis was P¼
n 0:5 N
while this estimator tends to slightly overestimate the Weibull modulus, it has been shown to give the least biased estimate of the modulus for sample sizes N P 20 [21,22]. The observed strength distributions for the four polysilicon conditions are shown in Fig. 5. The characteristic strength rh and Weibull modulus m for each of the four conditions was estimated in Table 3 using a two-parameter Weibull distribution for the probability of failure P as a function of applied stress r. m r P ¼ 1 exp rh 3.3. Surface topography Scanning electron microscopy revealed very distinct surface topography for the coarse-grained and fine-grained
443
Fig. 5. Observed strength distributions for each of the four Poly1 variants. Tests were conducted in air at room temperature. The error bars are not visible since the stress resolution (±2.5 MPa) was far smaller than the width of the data points (50 MPa).
variants, as shown in Fig. 6. The coarse-grained wafers exhibited well-defined grain boundary grooves that encircle some but not all of the grains on the top-surface. Also present are small (<50 nm) nodules or bumps. These bumps were previously thought to be SiC particles that result from migration and precipitation of residual carbon from the tetraethylorthosilicate (TEOS) sacrificial oxide during thermal annealing [23]. However, in the present study either undoped silane glass or 2% PSG was used in lieu of TEOS and there was no carbon source for the formation of SiC particles. We speculate that the nodules or bumps may instead be the nucleus of new grains initiating on the surface. The top-surface grain boundary grooves appear to be the deepest and sharpest notch-like features on the surfaces of the coarse-grained material and thus are expected to be the critical flaws that drive failure. In the case of the finegrained laminate structures, the top-surface is considerably smoother with no obvious preferential grain-boundary etching or nodular features. The much finer top-surface is not simply due to the smaller grain size: the average grain diameter of the fine-grained laminate is 125 nm whereas the top-surface features in the fine-grained laminate structures are much finer than 125 nm.
Table 2 Grain size measures (average ± 1 standard deviation). Wafer
Coarse-graine 2.0% PSG
Fine-grained 2.0% PSG
Coarse-grained undoped
Fine-grained undoped
Major diameter (nm) Minor diameter (nm) Avg. diameter (nm)
408 ± 227 160 ± 62 284 ± 137
153 ± 46 96 ± 24 125 ± 33
398 ± 184 181 ± 63 290 ± 115
158 ± 48 97 ± 38 128 ± 37
444
B.L. Boyce et al. / Acta Materialia 58 (2010) 439–448
Table 3 Weibull fit parameters for the four conditions. Wafer
Coarse-grained 2.0% PSG
Fine-grained 2.0% PSG
Coarse-grained undoped
Fine-grained undoped
Characteristic Strength, rh (GPa) Weibull modulus, m
1.48 8.7
2.83 13.5
1.76 12.9
2.80 8.6
Fig. 6. Oblique (45°) SEM view of the sidewall and top-surface of the four Poly1 variants. In Wafers 6 and 10, some top-surface grain boundaries are encircled by grain boundary grooves caused by preferential etching. Sidewall or corner notches indicated by arrows in were seen consistently in the heavily doped coarse-grained material and were not as extensive in the undoped coarse-grained material.
The differences in top-surface morphology and roughness were examined quantitatively by contact-mode atomic force microscopy (AFM) using a Park Scientific AutoProbe LS AFM with a 1–3 nm radius silicon stylus. For each of the four polySi variants, three to four 10 lm 10 lm regions were surveyed with a 256 256 pixel grid. Within each region, smaller 5 lm 5 lm and 2 lm 2 lm subregions were surveyed for increased resolution. Typical AFM images from the 5 lm scans are shown in Fig. 7. The distinction between the coarse-grained and fine-grained laminate variants is clearly consistent with the SEM images of Fig. 6. The resulting root-mean-squared (rms) surface roughness metric for each variant is shown in Table 4. The fine-grained laminate variants had 20% lower RMS roughness than the coarse-grained variants, consistent with the observation that the fine-grained laminates were stronger than the coarse-grained variants. The effect of phosphorous doping on the resulting RMS roughness was less significant and possibly even negligible compared to the effect of the grain size/structure. However, fracture strength is not directly related to average roughness over a field, but rather to the worst-case flaws in the material. An attempt was made to identify worst-case flaws in the surveyed regions, by obtaining high resolution images of the deepest features in the 10 lm regions surveyed. However, the observed “deepest” trenches, which were 20– 50 nm in depth, did not show any notable correlation with the strengths of the four variants. This apparent lack of correlation may be due to insufficient sampling. While AFM examination of the top-surface did not reveal any clear evidence of a doping-dependent change in flaw morphology, Fig. 6 shows that the coarse-grained 2.0% PSG
variant contained significant corner and sidewall defects (arrows in Fig. 6) that were not observed in the coarsegrained undoped variant. This qualitative observation is consistent with the mechanical strength in that undoped coarse-grained material was stronger than the doped coarse-grained material. In both the coarse-grained and fine-grained laminate materials, the sidewall edge appeared to be smoother than the top-surface. The sidewalls of the coarse-grained material exhibited vertical waviness or “curtaining”. In the heavily doped coarse-grained wafer, the curtains resulted occasionally in visible notch-like features at the top corner where grain boundary grooves intersected the sidewall (arrows in Fig. 6). The sidewalls of the fine-grained laminate material were notably smoother with less significant curtaining and no visible flaws compared to the coarsegrained material. 3.4. Fractography Typical fracture surfaces for the fine-grained laminate and coarse-grained materials are shown in Fig. 8. The observed fractography was consistent with the previous descriptions of microstructure and surface flaw topography. The coarse-grained material exhibited coarse crystallographic faceting corresponding to the larger grain size (i.e. only a few grains through-thickness). By comparison, the fracture surface of the fine-grained laminate exhibited much finer surface features with some horizontal delineation of the five-layer lamellar microstructure visible in the fracture surfaces. The fracture surface in Fig. 8b was tilted at an angle away from the expected mode-I plane, suggesting
B.L. Boyce et al. / Acta Materialia 58 (2010) 439–448
445
Fig. 7. Surface morphology of the four Poly1 variants as measured by atomic force microscopy.
Table 4 Surface roughness as measured by AFM. Wafer
6 coarse-grained 2.0% PSG
8 fine-grained 2.0% PSG
10 coarse-grained undoped
12 fine-grained undoped
5 lm 5 lm regions Surveyed RMS roughness (nm)
3 8.06 ± 0.11
4 6.92 ± 0.31
3 8.26 ± 0.44
4 6.22 ± 0.15
the possibility of some bending forces superimposed on the tensile forces. Crack initiation in brittle materials is often surmised from either obvious defects found on the fracture surface or from tracing back characteristic mirror–mist–hackle markings to the origin. In these micro-scale fracture surfaces, identifying failure origins is particularly challenging because: (a) traditional mirror–mist–hackle markings often do not develop over such a small scale, (b) the expected nano-scale defects (10–100 nm) are difficult to image by electron optics in a poorly conductive medium and (c) the crystallographic faceting is of a similar scale to the features of interest, thereby convoluting interpretation. Nevertheless, in the coarse-grained fractograph shown in Fig. 8a, the critical defect is likely the deep trench near the right sidewall of the specimen. The trench appears to be caused by excessive preferential etching of the grain boundary. In
the fine-grained laminate structure, no obvious defect was identified in the fracture surface. 4. Discussion 4.1. The role of grain boundary grooves The critical strength-limiting defects in the coarse-grained material appear to be grooves caused by preferential etching of grain boundaries. These grain boundary grooves are typical for common polySi processes. Miller et al. [16] has shown that the grain boundaries of MUMPs polySi appear to be a preferred path for structural degradation. When galvanic corrosion is induced by etching a multi-layer structure that is in electrical contact with gold, preferential grainboundary attack appears to be responsible for a transition from transgranular fracture to intergranular fracture.
446
B.L. Boyce et al. / Acta Materialia 58 (2010) 439–448
process shown in Fig. 10 confirm the direct correspondence between observed surface grooves and underlying grain boundaries. However, not all grain boundaries are attacked equally. Clearly, in the coarse-grained custom material of the present study, the grooves appear to outline only certain grains. Crystallographic texture analysis is ongoing to investigate the possible correlation between the deepest grooves and grain boundaries with the highest surface energy. While grain boundary grooves appear to be the sharpest, deepest notch-like feature in the coarse-grained material, it is not clear that they are responsible for failure in the fine-grained material. Nevertheless, the absence of well-delineated grain boundary grooves in the fine-grained laminate appears to enhance the resulting tensile properties: the characteristic fracture strength for the fine-grained laminate was 60–90% higher than that for the coarse-grained material. 4.2. The role of phosphorous doping
Fig. 8. Typical fracture surfaces for coarse-grained and fine-grained laminate materials.
SUMMiT polysilicon also shows evidence of preferential grain-boundary attack, as is evident from the grooves shown in Fig. 9. Cross-sectional TEM images from the SUMMiT VTM
The addition of phosphorous (n-type) doping lowered the average fracture strength of the coarse-grained polysilicon by 16%. This is a similar result to that of Ballarini [1] where boron (p-type) doping was found to lower the fracture strength by 12% compared to undoped material. Interestingly, the phosphorous doping had little or no effect on the fine-grained laminate material. The important distinction here is that the coarse-grained material showed clear grain boundary grooves at the top-surface whereas the smoother topography of the fine-grained material did not clearly correlate with grain boundaries. This suggests that the diffusion-driven dopant content in the present study does not likely alter the intrinsic toughness of the material [4], which would have affected both microstructures equally, but more likely, the dopant may exacerbate the localized grooving process. It is clear from Fig. 6 that the heavily doped coarse-grain material exhibited more pronounced crevices at the corners and sidewall surfaces than the undoped counterpart. 4.3. Statistical variability in fracture strength: implications for reliability
Fig. 9. In this fractured SUMMiT V polysilicon tensile bar, grain boundary grooves are visible both on the top-surface and on the sidewalls.
The Weibull modulus, determined for each of the four silicon variants in Table 3, describes the breadth of the statistical distribution in failure strengths: low values of Weibull modulus correspond to high degrees of scatter. Not only does the doped fine-grained laminate material have the highest characteristic strength of 2.83 GPa, but it also has the highest Weibull modulus value corresponding to the tightest distribution in fracture strengths. This has important implications for yield and reliability. For example, if an acceptable probability of failure was 1%, the doped fine-grained laminate material with a Weibull modulus of 13.5 would have an allowable design stress of 2.01 GPa or 71% of its 2.83 GPa characteristic strength. By comparison, the doped coarse-grained material with a Weibull modulus of 8.7 would have an allowable stress of
B.L. Boyce et al. / Acta Materialia 58 (2010) 439–448
447
Fig. 10. (left image) Annular dark-field TEM image of a sidewall surface crevice penetrating at a grain boundary and (right image) higher resolution bright-field TEM image of another sidewall crevice penetrating at a grain boundary. Some grain boundaries showed no signs of preferential attack.
0.872 GPa or 59% of its 1.48 GPa characteristic strength. For reliable design, the statistical scatter in fracture strengths can be as important as the average or characteristic strength of the material. The statistical scatter is thought to be related to the homogeneity of the flaw population, and possibly the homogeneity in the intrinsic fracture toughness. A more detailed discussion of this issue and its implications is presented in Ref. [24]. 4.4. Other considerations: role of environment Silicon fracture also involves an important environmental aspect that is not clearly understood. For example, Alan et al. [25] has shown that the strength of single-crystal Si with hydrogen-terminated surfaces degrades by as much as 30% during the first month of air exposure after release. However, microfabricated single-crystal Si that has been coated with a methyl monolayer is initially stronger than H-terminated single-crystal Si and shows no degradation after a month of exposure to air. These observations may be attributed to stress–corrosion cracking that can occur in stressed surface oxide films [26], which may form easier on H-terminated surfaces than on monolayer-protected surfaces. However, this mechanism is somewhat speculative and does not clearly identify the precise mechanism by which the strength is altered (i.e. by altering the defect morphology or the local resistance to fracture). 4.5. Optimizing the mechanical performance of silicon MEMS The present study illustrates the significant role that processing has on the resulting strength of polycrystalline silicon MEMS. By merely switching from a single-layer deposition process to a multi-layer laminate process, the strength of silicon was increased by a factor of two. However, we have by no means optimized the strength of silicon. Further, while we see that the processing conditions can affect the location, shape, and size of failure-critical
defects, we still have only very limited understanding of the processing–structure relationships that govern the formation of critical defects. While the present study has focused solely on the tensile strength of polySi, there are many other mechanical properties that can be relevant to MEMS components, including Young’s modulus, fracture toughness, fatigue resistance, wear resistance, etc. There is a continuing need for improved understanding of the role that processing plays on the microstructure and resulting properties of MEMS materials. As MEMS technology continues to evolve from its microelectronics origins, structural optimization studies for enhanced mechanical performance will enable a broader range of highreliability MEMS applications. 5. Summary and conclusions In this study, we observed the fracture strength distributions of phosphorous diffusion-doped and undoped polycrystalline silicon with either a 285-nm grained equiaxed microstructure or a 125-nm grained laminate microstructure. The coarse-grained materials exhibited significantly lower characteristic strengths (1.48–1.76 GPa) compared to the fine-grained laminate material (2.80–2.83 GPa). The difference in strength was attributed largely to preferential etching of grain boundary grooves that were considerably more pronounced in the coarse-grained material. The lack of significant grain-boundary grooving in the fine-grained laminate material may be related to the alternating deposition process which incorporated an “amorphous” deposition step. The presence of phosphorous doping lowered the characteristic strength of coarsegrained material by merely 16% and had little or no effect on the fine-grained material. Acknowledgements The authors would like to thank Dr L. Phinney for collaborating on the fabrication of these custom polycrystal-
448
B.L. Boyce et al. / Acta Materialia 58 (2010) 439–448
line silicon materials. The assistance of T. Crenshaw and B. McKenzie is greatly appreciated in support of mechanical testing and SEM imaging, respectively. BLB would like to thank Drs E.D. Reedy Jr., J. Foulk, M.P De Boer, T.E. Buchheit and J.S. Custer for useful discussions on this topic. Sandia is a multiprogram laboratory operated by Sandia Corporation, a Lockheed Martin Company, for the United States Department of Energy’s National Nuclear Security Administration under contract DE-AC0494AL85000. References [1] Ballarini R, Kahn H, Tayebi N, Heuer AH. Effects of microstructure on the strength and fracture toughness of polysilicon: a wafer level testing approach. In: Muhlstein CLBSB, editor. Symposium on mechanical properties of structural films. Orlando, (FL): American Society Testing and Materials; 2000. p. 37. [2] Cho SW, Jonnalagadda K, Chasiotis I. Fatigue Fract Eng Mater Struct 2007;30:21. [3] Foulk JW, Johnson GC, Klein PA, Ritchie RO. J Mech Phys Solids 2008;56:2381. [4] Cook RF. J Mater Sci 2006;41:841. [5] Son D, Kim JJ, Lim TW, Kwon D. Thin Solid Films 2004;468:167. [6] Ballarini R, Mullen RL, Heuer AH. Int J Fract 1999;95:19.
[7] Alsem DH, Boyce BL, Stach EA, Ritchie RO. Sens Actuator A— Phys 2008;147:553. [8] Chasiotis I, Knauss WG. J Mech Phys Solids 2003;51:1533. [9] Robertson WM. J Am Ceram Soc 1981;64:9. [10] Alan T, Hines MA, Zehnder AT. Appl Phys Lett 2006:89. [11] Miller DC, Boyce BL, Dugger MT, Buchheit TE, Gall K. Sensors Actuators: A Phys 2007;138:130. [12] Kahn H, Deeb C, Chasiotis I, Heuer AH. J Microelectromech Syst 2005;14:914. [13] Pierron ON, Macdonald DD, Muhlstein CL. Appl Phys Lett 2005:86. [14] Miller DC, Becker CR, Stoldt CR. J Electrochem Soc 2008;155:F253. [15] Miller DC, Gall K, Stoldt CR. Electrochem Solid State Lett 2005;8:G223. [16] Miller DC, Boyce BL, Gall K, Stoldt CR. Appl Phys Lett 2007:90. [17] Miller DC, Boyce BL, Kotula PG, Stoldt CR. J Appl Phys 2008:103. [18] McAllister PV, Cutler IB. J Am Ceram Soc 1972;55:351. [19] Sniegowski JJ, de Boer MP. Annu Rev Mater Sci 2000;30:299. [20] Boyce BL, Grazier JM, Buchheit TE, Shaw MJ. J Microelectromech Syst 2007;16:179. [21] Sullivan JD, Lauzon PH. J Mater Sci Lett 1986;5:1245. [22] Bergman B. J Mater Sci Lett 1984;3:689. [23] DelRio FW, Dunn ML, Boyce BL, Corwin AD, de Boer MP. T J Appl Phys 2006:99. [24] Boyce BL, Ballarini R, Chasiotis I. J Micromech Microeng 2008:18. [25] Alan T, Zehnder AT, Sengupta D, Hines MA. Appl Phys Lett 2006:89. [26] Kahn H, Avishai A, Ballarini R, Heuer AH. Scripta Mater 2008;59:912.