Strain engineering has been a key factor in the continued development of Si-based electronics from the early 2000s era onward. 1 Biaxial 2 and later uniaxial strain 3,4 in Si was used to enhance electron and hole mobilities in individual transistors to overcome challenges in transistor scaling. 5 Early methods developed along these lines used process-induced strain engineering techniques that resulted from the effects of the nanofabrication process itself, resulting in individually tailored strain for each transistor channel. 6 These techniques involved using embedded SiGe source-drain contacts through selective epitaxy, but also utilized stress transfer techniques through the deposition of stressed thin film liners that would transfer stress (and, therefore, strain) into the silicon channel. These techniques have large advantages in engineering transistor mobility since they can be applied on a device-by-device level and are highly scalable since they are equally applicable on the nanoscale as on the microscale. Particularly, since electrons and holes have different strain requirements for mobility enhancement, it was critical that process-induced strain engineering techniques could selectively apply uniaxial tensile or uniaxial compressive strain depending on whether the transistor was n-channel (NMOS) or p-channel (PMOS).
In this work, we explore the time and low-temperature stability associated with process-induced strain engineering with 2D materials, especially strain engineering through thin film stressor deposition onto 2D materials. This is borrowing directly from the idea of using of thin film stress liners from early commercial strain engineering processes. 7,8 In our previous work on this topic, it has been shown that through a simple evaporation step of a stressed thin film deposited directly onto 2D MoS 2 , that strain can be effectively transferred into few-layer MoS 2 on conventional substrates and single-layer MoS 2 when fabricated on h-BN. [9][10][11][12] Through this method, we have been able to engineer compressive or tensile strain through varying stressor compression or tension, engineer the magnitude of strain through varying stressor film force [film stress (σ)* film thickness(t)], engineer uniaxial or biaxial strain through geometric patterning, and engineer strain relative to the 2D crystal structure through selective patterning. One critical question that remains is whether these techniques are equally valid at all temperatures and whether they are stable in the long-term. Many interesting strain tunable properties of 2D materials exist at temperatures below room temperature, where effects like structural, electronic, optical, superconducting, or topological phase transitions may be activated via strain engineering at low temperature. 13 Similarly, proving the long-term stability of these techniques will significantly contribute to the eventual applicability of such techniques to potential electronic, optoelectronic, or quantum electronic applications in the future.
To explore the time stability of thin film stressor-induced strain engineering, we chose the model system of MoS 2 , which has been used in our previous work on the topic due to its well-studied strain-sensitive Raman modes. Strain is examined in a 3L-MoS 2 sample that has been directly exfoliated onto an MgO substrate [Fig. 1(a)]. Care has been taken to assure proper substrate adhesion, including ensuring that the surface roughness of the MgO substrate before exfoliation is low (R a ∼ 0.15 nm), pre-annealing MgO substrates at 150 °C in an O 2 and H 2 O controlled glovebox (<0.5 ppm each), 14 and exfoliating within the glovebox itself. Subsequently, the samples are subjected to an ultrasonic bath in acetone (30 min) and isopropanol (30 min) to verify adhesion. All poorly adhered flakes 2g peak position within 24 h of stressor deposition. 11 (f ) Raman spectroscopic mapping of E 1 2g peak position after 14 months from stressor deposition. Both Raman maps follow the color scale on the right of (f ). (g) Constructed image difference between (e) and (f ), map follows the pink/green color scale presented to the right. We note that there are potentially spatially correlated changes in the sample; however, these changes are close to the noise level in the measurement.
come off the substrate during the ultrasonic cleaning step. Following exfoliation, direct write laser photolithography and lift-off was used to pattern an e-beam evaporation-deposited stressor layer, which is a thin film that consists of 10 nm Al 2 O 3 /100 nm MgF 2 /10 nm Al 2 O 3 . The bottom layer of Al 2 O 3 is to enhance adhesion to the 2D material, while the top Al 2 O 3 layer prevents the stressor itself from relaxing due to exposure to environmental humidity (known to happen in thin film MgF 2 ). 15 MgF 2 is otherwise a material that has high tensile stress when deposited through evaporation due to low adatom mobility when deposited at room temperature. 15,16 Since almost all vacuum deposited materials contain thin film stress, the choice of material may be varied for compression or tension. 17,18 We have previously shown that strain transfer is directly proportional to film force [film stress (GPa) × film thickness (nm)], and in this case, the film force was 25 N/m as measured through the standard measurement of wafer curvature and the Stoney equation. 19,20 We have previously shown that by lithographically patterning the stressor layer, geometry allows for engineered uniaxial or biaxial strain to be transferred 11 [Fig. 1(b)]. This can easily be imagined for a stressor with tensile stress, as the stressor itself wishes to relieve stress by shrinking (transferring compression). At the edges, there is a unidirectional pulling force (uniaxial tensile strain), while at the center, there is a bidirectional compressive force (biaxial compression). An optical micrograph of the final device is shown in Fig. 1(c) and is the exact same device in our previously published work on the topic. 11 We use Raman spectroscopic mapping (WiTec Alpha300R Confocal Raman microscope, 532 nm excitation laser, ≤0.75 mW power, ∼0.7 μm spot size, and 1800 l/mm spectrometer grating) to determine the strain state in our samples.
In the MoS 2 trilayer sample with a lithographically patterned stressor layer, it is shown to vary from tensile to compressive strain depending on the distance from the edge. Figure 1(d) shows individual Raman spectroscopic traces for three regions labeled A (center of stressor), B (edge of stressor), and C (control). The E 1 2g peak (∼384 cm -1 ) is most sensitive to in-plane strain and is shifted in opposite directions relative to the control, indicating both compressive (point A) and tensile (point B) strain transfer. The A 1g peak (∼407 cm -1 ) position does not change as much, since this peak is less sensitive to in-plane strain. A clearer picture is shown through an E 1 2g peak position map of the entire flake structure, with blue regions representing tensile strain and red regions indicating compressive strain relative to the white control [Fig. 1(e)]. We compare this same Raman spectroscopic map in Fig. 1(e), which was taken less than 24 h after stressor deposition, to a map taken 14 months after deposition noting that the strain distribution pattern and magnitude are qualitatively alike over this time period [Fig. 1(f)]. We can quantitatively analyze changes in strain with time by comparing peak position change over the 14-month period at specific points. Here, we compare the change of E 1 2g in regions A and B (relative to region C) with time, with changes of +0.842 ± 0.13 and -0.735 ± 0.12 cm -1 before, to +0.811 ± 0.13 and -0.696 ± 0.12 cm -1 after. The margins of error we present here are that from Lorentzian peak fittings; therefore, this uncertainty derives from noise in the optical signals. We also note that these magnitudes in Raman shift are consistent with the top layer being strained ∼0.85%, and the overall measured magnitude is reduced in a 3L-MoS 2 due to incomplete strain transfer in the c-axis. The out-of-plane strain transfer length scale in MoS 2 has been found to be approximately two layers [Fig. 1(a)], where this parameter will vary with the strength of the specific 2D system's interlayer coupling. 9,10 The minimal change in Raman shifts with time show that we retain this 0.85% strain within the devices with a maximal change of strain around 0.05%. Upon extracting a difference map between the 14-month measurement and the <24-h measurement, we confirm the shifts to be within error of ±0.12 cm -1 [Fig. 1(g)]. This shows that for lithographically patterned stressors on exfoliated 2D materials, the process is long-term stable; although the ultimate limit of time stability has not been established, it is likely as long as commercial process-induced strain engineering techniques, which have been in use since 2003.
Next, we explore the same concept of time stability on strain engineered van der Waals (vdW) heterostructures. We have shown in previous studies that the last layer of MoS 2 is well adhered to the substrate when directly exfoliated and resists strain transfer from the same strain engineering techniques we introduced prior. This can be alleviated by placing 1L-MoS 2 on h-BN, where weak van der Waals bonding between the layers allows for layer sliding and the ability to strain engineer the 1L-MoS 2 using process-induced stress. 10 Here, we construct a heterostructure of 1L-MoS 2 /h-BN on SiO 2 /Si (top layer of MoS 2 is exposed) using glovebox-enclosed standard dry transfer van der Waals heteroepitaxy techniques [Fig. 2(a)]. Instead of lithographically patterning the stressor layers, we uniformly encapsulate the devices with a thermally evaporated stressor consisting of 5 nm CrO x /100 nm MgF 2 (20 N/m), where the top CrO x layer is used to prevent stress relaxation in MgF 2 . Uniform encapsulation has been shown to apply uniform biaxial compression to the underlying MoS 2 layer, since tensilely stressed films wish to uniformly relax the stress through compression [Fig. 2(b)]. Figure 2(c) shows an optical micrograph of the fully encapsulated heterostructure device and represents the exact same device as presented in our previous works on strain engineering van der Waals heterostructures. 10 Using the same techniques as in the previous section, we analyze strain using Raman spectroscopic mapping, examining before and after stressor encapsulation we can extract an average change in the in-plane-sensitive E 1 2g peak of +2.46 cm -1 , indicating 0.47% biaxial compressive strain [Fig. 2(d)]. Upon also comparing the E 1 2g peak position maps before and after (<24 h after stressor deposition) encapsulation, there is a uniform increase in E 1 2g peak position. This indicates the uniform transfer of biaxial compressive strain through the entire MoS 2 monolayer on h-BN. Examining the same structure 4 months post-deposition leads to the map presented in Fig. 2(g), where there are no noticeable changes. Quantitatively, the change in E 1 2g peak position (relative to the before E 1 2g ) was 2.46 ± 0.23 cm -1 , and after 4 months, it is 2.39 ± 0.22 cm -1 ; we use a 521 cm -1 Si peak for calibration/accuracy of these changes. We have retained the ∼0.47% biaxial compressive strain within error, over the course of the observation period. We again construct a difference map between the <24 h sample and the 4-month Raman maps of this sample and confirm
only changes within the noise level are observed [Fig. 2(h)]. Therefore, from our measurements, we know that this technique is time stable for van der Waals heterostructures and monolayer 2D materials as well.
Finally, we examine our strain engineering process as a function of temperature. Strain has been shown to modify the MoS 2 bandgap, 21,22 which can be detected through changes in the exitonic behavior by photoluminescence (PL) spectroscopy (532 nm excitation laser, 0.1 mW power, ∼0.5 μm spot size, and 150 l/mm spectrometer grating). Here, two van der Waals heterostructures of 1L-MoS 2 /h-BN are fabricated and measured, but only one is strained with the stressor from the previous section. When room temperature PL is compared between the two samples, the compressive biaxially strained sample has an A exciton peak shift of 38 meV, which corroborates the extracted ∼0.46% biaxial compressive strain that was determined through Raman spectroscopy in the previous section [Figs. 3(a) and 3(b)]. In Figs. 3(a) and 3(b), we also present the full temperature-dependent PL of both the control heterostructure and the strained heterostructure, respectively, as a function of temperature. To judge whether strain is retained at low temperature, we can look at the A exciton peak position in both sample cases [Fig. 3(c)]. It is shown that for the strained sample, the A exciton peak position is always ∼38 ± 5 meV above the control value at all temperatures, suggesting that strain is both transferred at room temperature and also retained at low temperature down to 4 K. We show here that the stressor layers
themselves are immune to low-temperature cracking arising from changes due to differential thermal contraction at low temperatures. This result fits with our model since the thin film stressors relax to impart strain into the 2D material to relieve stress after the deposition process. Temperature-dependent changes due to thermal expansion or contraction are small relative to the process-induced stress in the films, and we can calculate ∼0.045% strain introduced from thermal expansion coefficient mismatching between 1L-MoS 2 and thin film MgF 2 from 293 to 4 K. 23,24 This is an order of magnitude smaller in comparison to the 0.47% biaxial compressive strain from the effect of stressor-induced strain onto the MoS 2 monolayer. The observed small variations (∼4 meV) in the shifted PL peak position in the encapsulated samples as a function of temperature may arise from these smaller thermal expansion contributions. We note that we only calculate differential thermal mismatch at the MgF 2 /1L-MoS 2 interface, since this is the only interface that differentiates the strained sample from the measured control sample.
While the low-temperature peak of the neutral A exciton changed with respect to strain, we can also observe in the encapsulated sample that the trion contribution to the PL signal is diminished relative to the control. Temperature-dependent intensity of the trion peak to the neutral A exciton peak is shown in Fig. 3(d). This effect can be explained by local charge transfer from the MgF 2 layer, leading the excess electrons in the 1L-MoS 2 to be reduced. We can quantitatively estimate this by looking at a ΔE 1 2g vs ΔA 1g plot of the strained monolayer sample before and after deposition [Fig. 3(e)]. A vector space can be created via a linear transformation of each phonon's Grüneisen parameter, and to independently quantify strain (ε) and electron concentration (n), this an effective method commonly employed to decouple strain and doping in both graphene and MoS 2 . 25,26 Utilizing this vector space, we can
observe ΔE 1 2g vs ΔA 1g from before and after adding the stressor and then extract the electron concentration that is roughly reduced by ∼0.1 × 10 13 cm -2 . This is not an indication of layer damage, as we have previously shown that the stressor application is a damage-free process by examining the homogeneous broadening effects of the Raman peaks in this sample, showing to the limit of our detection that the damage is minimal. 10 Instead, this is likely due to charge transfer from the dielectric itself, which has been shown in many works to p-dope the MoS 2 layer leading to trion reduction. [27][28][29][30] We have also compared our stressor encapsulated vdW heterostructure to a similar heterostructure encapsulated in only thermally evaporated 100 nm MgF 2 . Since the top layer is not sealed with a 5 nm CrO x layer, the stress in the MgF 2 layer changes with exposure to humidity in ambient air. Here, the heterostructure shows no strain transfer, but also shows the same relative reduction in trion intensity [inset of Fig. 3(d)].
We have shown through the Raman and photoluminescence spectroscopic analysis that process-induced strain engineering techniques on 2D materials can endure low temperatures down to 4 K and are time stable up to 14 months. These values are likely not the ultimate limit of the temperature range of these stressor techniques, as thermal expansion at lower temperatures is minimal below 4 K, and stressor-driven strain engineering techniques in industry have been commercially in use in silicon electronics since the 90 nm technology node in 2003. This study opens the door to any number of interesting strain tunable 2D material properties such as superconductivity or single-photon emission in engineered TMDCs that exist at low temperatures. Since process-induced strain engineering techniques have a long history of scalability and have now been shown to be time stable in 2D materials, it implies that any of these strain-engineered 2D systems may have direct impact on densely integrated (opto)electronic applications at all temperatures.
J. Appl. Phys. 131, 024304 (2022); doi: 10.1063/5.0075917