1. Introduction

Membrane structures have attracted ample interest over the years from many disciplines with applications in micro-electro-mechanical systems (MEMS) [1], cavity opto-mechanics [2–4], and optical lasing [5,6]. Depending on the application, diverse membrane materials are used with most applications resorting to silicon (Si) or dielectrics such as silicon nitride (Si₃N₄) and silicon dioxide (SiO₂) [7–11]. The fabrication process differs for each. Nevertheless, the processes are established and standard membrane types are commercially available. However, applications pertaining to lasing require intricately patterned membranes and typically employ alloyed semiconductor gain material such as InGaAsP with quantum wells [12]. These membranes are periodically patterned into photonic crystals (PhCs) to highly confine light of certain wavelengths or, in other words, sustain high quality factor (Q) modes required for lasing [13,14]. Thus, these membrane PhCs serve as optical cavities for lasers. Here, we construct photonic crystal cavities with InGaAsP that employ bound states in the continuum (BIC) for lasing [15–19]. Bound states in the continuum (BICs) are waves that exist within a continuum of radiating waves and yet do not radiate or decay. Contrary to conventional wisdom, these states remain localized or bound to the cavity [20].

Unlike conventional PhCs that most often take the complimentary form, i.e. periodic array of holes, these suspended PhC cavities are composed of periodically spaced and interconnected cylinders. Furthermore, because the cylinders’ radii are critical to the lasing mode in BIC lasers, stringent requirement is placed on fabrication precision. Hence, the fabrication of these new devices offers its own unique challenges which include maintaining pattern dimensionality while realizing a fully suspended and mechanically stable membranes. Here, we describe two different approaches of fabricating these BIC membrane lasers and their effect on device performance. One approach involves a hydrogen silsesquioxane (HSQ) electron beam lithography resist serving as a dry etch mask and another with an added chromium (Cr) hard mask. We find that the performance of these devices is sensitive to the fabrication quality and is hampered by deviations in the device dimensions from those intended. Moreover, we elaborate on dry etch requirements in conjunction with the optimization of etching window geometry for a quick membrane release by selectively wet etching the substrate. We investigate etch windows that have either rectangular or trapezoidal openings which offer different etch overlaps among crystal planes of the InP substrate. An etch window that has the maximum etch overlap results in the quickest membrane release given the same dry etch depth.

2. Device fabrication

In this work, BIC membrane lasers are fabricated using standard nanofabrication techniques. We choose the active medium to be epitaxially grown multiple quantum wells of InGaAsP material lattice-matched to InP substrate and tailored to emit in the telecommunication band (1.5-1.6 μm). The gain material consists of nine In_{x = 0.564}Ga_1-xAs_{y = 0.933}P_1-y quantum well layers of 10 nm thickness (bandgap of 1.6 μm) and In_{x = 0.737}Ga_1-xAs_{y = 0.569}P_1-y barrier layers of 20 nm thickness (bandgap of 1.3 μm). An additional top barrier layer of 30 nm makes the total height of the gain 300 nm (Fig. 1(a)). All finished samples are composed of a system of periodic cylindrical resonators interconnected by bridges for mechanical stability (Fig. 1(b)-1(c)). The patterns including the interconnecting bridges are defined by dry etching with the help of a hard etch mask. In what follows, we employ two fabrication approaches each with different etch masks to realize these structures. One (Method I) involves a HSQ electron beam lithography resist and another (Method II) with an added Cr hard mask. The etch resistance of the two masks differ drastically and thus directly affect the final dimensions of the patterns.

 

Fig. 1 (a) Schematic of epitaxially grown InGaAsP layers on InP substrate with nine quantum wells (in red). (b) Top view of laser membrane with cylindrical resonators arranged in a square lattice interconnected by bridges and a secondary outer pad. (c) Tilted view of the membrane with a magnified view of the cylindrical resonators with radius (R), thickness (H = 300 nm), and period (P = 1.2 μm).

Download Full Size | PDF

A. Method I (HSQ hard mask)

Figure 2 shows the process flow for the fabrication of BIC membrane lasers using a HSQ negative tone resist acting as the hard mask for reactive-ion etching (RIE). Starting with the epitaxially grown wafer (Fig. 2(a)), 6% HSQ in methyl isobutyl ketone (MIBK) is spin-coated at 2500 rpm for 60 s and soft-baked at 180 °C for 60 s to yield a film thickness of 130 nm. Subsequently, the HSQ is exposed at 100 kV and 3 nA beam current with 800 μC/cm² dose using a Vistec EBPG5200 electron beam lithography machine. The sample is developed using 25% tetramethylammonium hydroxide (TMAH) in water developer for 60 s (Fig. 2(b)). In step c, a purely RIE process is performed at a base pressure of 30 mTorr and a temperature of 35 °C with a RF power of 150 W using a Trion RIE/ICP Dry Etcher. The etch is carried out with 10 sccm (standard cubic centimeter per minute) of methane (CH₄) flow, 40 sccm of hydrogen (H₂), and 7 sccm of argon (Ar) combination for 680 s with an estimated etch rate of 70 nm/min to yield an etch depth of 800 nm [21]. Here, the etch time is chosen such that all of the 300nm of InGaAsP would be etched in addition to a considerable thickness of InP for easy membrane release. Next, residual organic contamination and polymer buildup during RIE are removed with a microwave oxygen (O₂) plasma treatment with an O₂ flow rate of 120 sccm (150 W) for 15 min. The HSQ layer is removed with 30 s of buffered oxide etchant (BOE) with a ratio of 6:1 (H₂O:HF) (Fig. 2(d)). Next, with the help of photolithography and a hydrochloric acid (HCl) based wet-etching solution, we remove a substantial amount of InP substrate below InGaAsP [22–24]. In step e, the areas to be wet-etched are opened in the negative-tone NR9-1500PY photoresist spun at 3500 rpm for 40 s to yield a thickness of 1.5 μm. After a 20 s UV exposure with the Karl Suss MA6 Mask Aligner and a reversal bake at 100 °C for 60 s, the resist is developed for 35 s with RD6 developer. Lastly, a diluted solution of hydrochloric acid (HCl:H₂O::3:1) with three parts acid to one part water by volume is used to selectively and anisotropically etch InP while minimally etching InGaAsP for a total etch time of 3 min (Fig. 2(f)-2(g)). Radii of the final cylindrical resonators are smaller than the radii defined after e-beam lithography mainly due to the eroding (i.e. narrowing) HSQ hard mask during RIE [25, 26]. Consequently, the InGaAsP sidewalls are also eroded. This reduction in dimensions is exacerbated with increased etch depth or etch time. The reduction in radii for an etch depth of 800 nm was ~70 nm. The reduction in dimensions also applies to the interconnecting bridges, and consequently, the mechanical stability of the membrane is drastically weakened. To compensate for this reduction, the dimensions can be over defined by the e-beam lithography mask. However, for a periodic structure the maximum radii of the cylinders defined by the lithography mask is limited to half the period (P/2). Hence, dimensions close to P/2 cannot be realized with the current fabrication process with HSQ as the etch mask. Therefore, a tougher etch mask is needed to preserve the dimensions defined by e-beam lithography during the dry etching process.

 

Fig. 2 Device fabrication process without a metal hard mask starting with the epitaxially grown multiple quantum wells on InP substrate and ending with the nanocylinders suspended membrane (a-g). Note that the bridges connecting the cylinders are intentionally drawn thinner as a guide to the eye. Both the bridges and the cylinders are of the same thickness.

Download Full Size | PDF

B. Method II (chrome hard mask)

To address the issue of reduced dimensions during the RIE step, we modify the above fabrication process to include a metal hard mask in addition to the HSQ. However, this process requires some initial preparation of the wafer (see Fig. 3). A 100 nm of 950 PMMA A2 is spin coated at 1500 rpm for 60 s and soft-baked at 180 °C for 60 s. Subsequently, 30 nm of Cr is thermally deposited on top at a rate of 0.4 Å/s using a Denton 502 Thermal Evaporator. The Cr layer serves as a dry etch hard mask to attain and preserve the critical dimensions of the membrane. The PMMA layer serves as a sacrificial layer for the final and easy removal of the Cr (Fig. 3(b)). The Cr is not directly deposited on the InGaAsP as it is extremely difficult to remove and adds significant losses to the optical device. Similar to method I, the wafer is spin-coated with 130 nm of HSQ resist. However, it is critical that the HSQ not be soft-baked at this step as the PMMA underneath will reflow and cause the Cr layer to crack extensively. Not soft-baking HSQ resist has been shown to avoid thermally induced contrast reduction [27, 28]. Hence, in comparison, not soft-baking slightly improves the contrast but does not significantly affect the cylinder definition. Next, the HSQ is exposed at 100 kV and 3 nA beam current with 800 μC/cm² dose and developed using 25% tetramethylammonium hydroxide (TMAH) in water developer for 60 s (Fig. 3(c)). In step d, RIE processes are performed using the Oxford Plasmalab 80 + to sequentially etch the Cr and PMMA layers. First, the Cr layer is etched with a base pressure of 90 mTorr and a temperature of 15 °C with a RF power of 30 W. The etch is carried out with a combination of 3 sccm of O₂ flow and 50 sccm of chlorine (Cl₂) for 7 min with an estimated etch rate for Cr being 10 nm/min. Note that the Cr is over-etched to ensure that all Cr residue on the surface of the PMMA is removed, as any remaining Cr will block the subsequent dry etching of the PMMA. Next, the PMMA layer is etched with a base pressure of 50 mTorr and a temperature of 20 °C with a RF power of 50 W. The etch is carried out with only 50 sccm of O₂ flow for 2 min 40 sec. It is worth noting that in addition to etching PMMA, the oxygen plasma extensively undercuts the PMMA below the Cr (black arrows in Fig. 3(d)). Thus, the etch needs to be tightly controlled such that all the PMMA is removed from the InGaAsP surface and yet the undercut is minimal [29]. The PMMA undercut ultimately contributes to rough sidewalls in the patterned InGaAsP. Both the Cr and PMMA dry etch steps were thoroughly optimized. Following the Cr/PMMA etch, we dry etch the InGaAsP/InP as described in method I except with a longer etch time of 16 min for a total etch depth of 1200 nm (Fig. 3(e)). This etch depth is deeper compared to method I (800 nm). Here, in contrast to method I, the higher etch resistance of the metal hard mask allows for a deeper etch into the InP substrate with minimal reduction in pattern dimensions. The deeper etch is required and is conducive for an easy membrane release discussed below. Next, the HSQ/Cr/PMMA stack is removed (Fig. 3(f)). Starting with top layer, HSQ is easily removed with the help of BOE (6:1). To lift-off Cr, the sample is submerged in acetone for 2 hours with slight sonication so all the PMMA is attacked and the Cr layer is lifted off. Following the lift-off, the same wet etching recipe described in method I is used to suspend the membranes (Fig. 2(e)-2(g)). However, in comparison with the sole HSQ hard mask, we now etch for a shorter 2 min 18 sec for complete membrane suspension due to the deeper dry etch of the III-V material. Ultimately, the reduction of cylinders’ radii is minimized with the use of the Cr metal mask.

 

Fig. 3 Device fabrication process involving a metal hard mask, starting with the epitaxially grown multiple quantum wells on InP substrate (a-f). The subsequent membrane release process is the same for both processes: with and without a metal hard mask (Fig. 2(e)-(g)). Note that the bridges connecting the cylinders are intentionally drawn thinner as a guide to the eye. Both the bridges and the cylinders are of the same thickness.

Download Full Size | PDF

3. Results and discussion

Figure 4 is a direct comparison of the dry etch quality of InGaAsP/InP with a HSQ hard mask (Method I) and an added Cr hard mask (Method II). A control pattern was etched for both cases while aiming for similar etch depths. With HSQ serving as the hard mask and its narrowing during the dry etch, the InGaAsP sidewalls have eroded inward directly resulting in the reduction of lateral dimensions as seen in Fig. 4(a) and in 4b with the HSQ removed. However, with the added Cr layer, the patterns etched in InGaAsP experience minimal reduction in the lateral dimensions as seen in Fig. 4(c) and in 4d with the mask removed. This comparison is further apparent from the analysis of bridge reduction of the finished devices shown in Fig. 5. In Fig. 4(c), the undercut in PMMA due to the oxygen plasma is clearly visible underneath the Cr layer. After the initial etch over the PMMA thickness, the undercut rate was estimated to be 20 nm/min [29]. A PMMA layer that is not undercut below Cr during its etch has rough edges and consequently this roughness is transferred to the InGaAsP layer thereby degrading the devices. On the other hand, a PMMA layer that is heavily undercut will once again lead to a reduction in lateral dimension, i.e. cylinder radius and bridge width. The dimension of the radius directly affects the cavity mode of the laser and the widths of the bridges affect the mechanical stability of the eventually suspended membrane.

 

Fig. 4 Electron micrograph images of Reactive Ion Etching (RIE) of InGaAsP/InP test patterns. Method I: Side view after dry etch with HSQ as etch mask (a) and after HSQ removal (b). Method II: Side view after dry etch with an added Cr hard mask (c) and after mask removal (d). Both were dry etched to a depth of 950 nm. The sidewall erosion is visible with a HSQ hard mask (a) compared to an added Cr hard mask (b). An undercut in the PMMA layer below the Cr is also observed.

Download Full Size | PDF

 

Fig. 5 Reduction in bridge widths, W, from the nominal bridge widths as a function of the nominal radius for (a) Method I: HSQ mask and (b) Method II: added Cr mask. Left inset is a schematic of cylindrical resonators of radius, R, with interconnected bridges with widths, W. Right insets are images of the finished devices. The vertical error bars are the standard error in the measurement of the bridge widths. With a HSQ mask, the bridge reduction is worse than with an added Cr mask. However, in both cases, the bridge reduction is greatest (lowest) when the cylinder radius is small (large). With small radii cylinders and for a fixed periodicity, there is more access to the sides of the bridges for the dry etch gases. This contributes to an increased sidewall erosion and thus thinner bridges.

Download Full Size | PDF

We show in Fig. 5 the reduction in bridge widths from their nominal or defined values as a function of the nominal radius of the cylinders for Method I (HSQ mask) and Method II (added Cr mask). The two processes have different original bridge widths, as seen in Table 1, since the bridges are expected to shrink dramatically for the HSQ process compared to the Cr. Hence, it is more appropriate to look at the ‘width reduction’ of the bridges from the original. With a HSQ mask, the bridge reduction is worse than with an added Cr mask. As seen in the image insets of Fig. 5, HSQ mask yields visibly thinner bridges. However, in both cases, the bridge reduction is greatest when the cylinder radii is small and lowest when the radii is large. With small radii cylinders and for a fixed periodicity, there is more access to the sides of the bridges for the dry etch gases. This contributes to an increased sidewall erosion and thus thinner bridges. Overall, the Cr mask and its increased etch resistance help alleviate the erosion in the membrane’s most critical elements: the bridges. The etch selectivity of HSQ is ~7-8 and in comparison, the selectivity for Cr is greater than ~50.

Table 1. Original dimensions of bridge widths and radii defined for HSQ and Cr processes

View Table | View all tables in this article

For method I, it is worth noting that the etching resistance of HSQ could not be improved significantly with an oxygen plasma post-treatment as previously suggested [30]. This is mainly due to the drastic variation in system conditions and in the gases used for the dry etch. As for method II, to alleviate the need for a PMMA spacer, selective wet etching of the Cr layer directly deposited on InGaAsP were explored including CR-7 (perchloric acid based) and CR-1020 (nitric acid based) etchants, however all etchants significantly attack and damage InGaAsP [31].

Following the dry etch of InGaAsP/InP material, the InGaAsP patterns are suspended by selectively removing the InP underneath with the help of HCl solution. The successful suspension of the InGaAsP patterns is contingent on the dry etch depth into InP and the geometry of the surrounding etching window as seen from Fig. 6. There is a strong crystallographic dependence on the etch rates of InP with HCl solution [22]. As seen in Fig. 6(a), the etch is halted by the indium (In)-rich {111} planes sloped at 55° from the plane of InGaAsP forming etch pits [23]. If the InP dry etch depth is shallow, adjoining {111} planes of neighboring openings do not meet and the suspension process is completely halted by the etch pits. Therefore, the total dry-etch depth (h) required for adjoining {111} planes to meet and fully suspend the InGaAsP membrane of thickness (t) is: $h\ge R\tan \left(55°\right)+t$ where R is the radius of the cylinders. Hence, it is imperative that the dry etch of InP be sufficiently deep in addition to the use of an appropriate etch window. As seen in Fig. 6(b) and listed in Table 2, we employ three different etching windows for the square patterns: i. Rectangular with opening widths (2P) and supporting arms (2.5P), ii. Trapezoidal with thick supporting arms (3P) and opening widths (4P), iii. Trapezoidal with thinner supporting arms (P) and opening widths (4P) where P = 1.2 μm [32]. All three etch windows lead to suspended structures given the appropriate dry etch depth is reached. The less efficient the geometry of the etch window, the more the dry etch depth required. In our case, the rectangular etch window is the least efficient. It is worth highlighting that a longer etch depth leads to a reduction in lateral dimensions even with a Cr metal mask. Therefore, it is prudent to optimize the geometry of the wet etch window for a quick membrane release. For a given geometry of the etch window such as the rectangular one, we see both a suspended array which subsequently collapsed (Fig. 6(c). i) and an etch-halted array (Fig. 6(c). ii). The two samples were processed together where both are dry-etched for an etch depth of 600 nm and subsequently wet-etched for 1 min 30 sec in HCl:H₂O (3:1) solution. The only difference being the cylindrical resonators in one array have smaller radii than the other. The array in Fig. 6(c). i with a measured radius of 440 nm, allows for larger dry-etched openings in-between the cylinders compared to the array in Fig. 6(c). ii with a larger radius of 540 nm. The larger openings make it easier for the neighboring {111}-InP planes to meet for a given dry etch depth. Hence, the smaller radii membrane is easily released whereas the other is halted by the formation of etch pits. It is worth noting that only a HSQ hard mask was used for these two samples. Therefore, the released membrane collapsed due to the reduction of the supporting bridge widths past their breaking point during the dry etch process. Similarly, wet etch of an array with a trapezoidal geometry is halted by the formation of etch pits due to thick supporting arms as seen in Fig. 6(d). Even prolonging the wet etch to 4 min did not break the etch pits. However, a trapezoidal etch window with thin supporting arms allows for a quick (2 min 18 sec) membrane release leaving a visibly large V-groove that runs underneath as seen in Fig. 6(e). Larger arrays were also fabricated and, as expected, required longer wet-etching times for complete suspension.

 

Fig. 6 Etch requirements and optimization of membrane release. (a) Dependence of InP dry-etch depth on the crystallographic selective wet-etching of InP. The etch is halted by the slowest set of etch planes, indium (In)-rich {111} planes of InP sloped at 55°. Therefore, the total dry-etch depth (h) required for adjoining {111} planes to meet and fully suspend the InGaAsP membrane of thickness (t) is: $h\ge R\tan \left(55°\right)+t$ (b) Three different wet-etch windows for membrane release: i. rectangular windows with opening widths (2P) in white and supporting arms (2.5P) in blue, both indicated by black arrows, ii. trapezoidal windows with thick supporting arms (3P) and trapezoidal opening widths (4P), iii. trapezoidal windows with thin supporting arms (P) and opening widths (4P) with P = 1.2 μm. (c-f) Electron micrograph images of completed fabrication for different etch windows and etch conditions. (c) Two 10x10 arrays with rectangular etching windows where both are dry-etched for an etch depth of 600 nm and subsequently wet-etched for 1 min 30 sec in HCl:H₂O (3:1) solution. (i.) Collapsed array with measured cylindrical radii of 440 nm after wet etching and (ii.) equivalent array with larger measured radii of 540 nm with halted etch due to formation of etch pits along {111} plane of InP. (d) Array with trapezoidal etch window with a halted etch due to thick supporting arms of InGaAsP indicated by outer white arrows and halted InP etch underneath indicated by second pair of white arrows. There are visible etch pits between the cylindrical resonators. All the etch pits remain even after a prolonged wet-etch of 4 min. (e) Successfully suspended array with trapezoidal etch window due to the thinner supporting arms (white arrows) which was wet-etched for a total of 2 min 18 sec. A visibly large V-groove runs underneath the fully released membrane along the {011} direction.

Download Full Size | PDF

Table 2. Summary of wet etch windows and their dimensions for membrane release

View Table | View all tables in this article

Both method I with a HSQ mask and method II with an added Cr metal-mask yield functional devices. However, the metal mask offers a tighter control over the finest dimensions of the pattern, i.e. radius and bridge widths (see Fig. 7). A completed 10x10 array with HSQ as the hard mask can be seen in Fig. 7a and a magnified view in Fig. 7(b). Similarly, a 10x10 array with Cr hard mask can be seen in Fig. 7(c) and a magnified view in Fig. 7(d). As seen, the cylinders fabricated with the Cr hard mask maintain the mask dimensions with straighter sidewalls whereas with the HSQ mask both the cylinders and bridges shrink drastically with sloped sidewalls. There is also some noticeable sidewall roughness when using a Cr/PMMA hard mask mainly due to the etch quality of the PMMA.

 

Fig. 7 Electron micrograph images of completed membrane structures with 10x10 cylindrical resonators interconnected by a network of bridges with a visible etch pit below in the InP substrate for HSQ etch mask (a) and added Cr hard mask (c). Respective zoom-in images of two cylinders at the center of the array (b, d). As seen, the cylinders fabricated with the Cr hard mask maintain the mask dimensions with straighter sidewalls whereas with the HSQ mask both the cylinders and bridges shrink drastically with sloped sidewalls.

Download Full Size | PDF

4. Device performance

In Fig. 8, we observe lasing from devices made using a HSQ hard mask and a Cr hard mask. In Fig. 8(a), we observe lasing from a fully suspended 10x10 array fabricated with a HSQ hard mask emitting at 1567 nm with a threshold power of 67 μW. Similarly, we also observe lasing from a 10x10 array fabricated with the help of a Cr hard mask but emitting at 1540 nm with a threshold power of 89 μW (Fig. 8(b)). Both arrays are optically pumped by a 1064 nm laser with 12 ns pulse width at a repetition rate of 300 kHz (Fig. 8(c)).

 

Fig. 8 Lasing from fully suspended devices fabricated using a HSQ hard mask (a) and a Cr hard mask (b). Both devices are 10x10 arrays optically pump at 1064 nm and operating at room temperature as seen by IR camera during testing (left insets). (a) Light-Light (LL) curve of the laser fabricated using a HSQ hard mask emitting at 1567 nm (right inset) with a threshold of 67 μW (vertical dotted line). (b) LL curve for a laser fabricated using a Cr hard mask emitting at 1540 nm (right inset) with a threshold of 89 μW. (c) Schematic of photoluminescence setup for the characterization of membrane lasers with a pulsed pump (1064 nm) path in blue and emission/imaging path in red with a CCD camera, a monochromator, and an InGaAs detector tied to a lock-in amplifier. A microscope objective (NA = 0.4) and L1 to L8 correspond to the lens assembly to the sample.

Download Full Size | PDF

By design and in virtue of the limitation of the two fabrication methods, the dimensions within the laser array (radius and bridge widths) are different which influences the lasing mode and thus dictates the lasing wavelength and threshold power. Hence, the two lasers emit at different wavelengths and have different thresholds. It is also worth noting that different etch windows were used for the two devices shown here. A rectangular etch window was used for the HSQ hard mask and a trapezoidal window with thin supporting arms for the Cr hard mask. The etch window has no effect on laser results but rather a considerable effect on the release of the lasing membranes during fabrication. The lasing mode is strictly dictated by the dimensions of the cylinders and bridges in the array. An optimized etch window such as the trapezoidal maximizes the etch overlap among crystal planes resulting in a quicker membrane release time for a given dry etch depth [32]. This also allows for a shorter dry etch time (or etch depth) thereby preserving dimensionality of the cylinders and bridges.

5. Concluding remarks

We have described two different approaches of fabricating suspended membrane lasers composed of periodic cylindrical nanoresonators on an InGaAsP/InP platform. One approach involves HSQ as a dry etch hard mask and the other uses a Cr metal hard mask. The HSQ mask leads to significant reduction in dimensions during the dry etching process. We observe that this reduction affects the mechanical stability of the membrane by eroding the bridges that keep the cylindrical resonators suspended. It also affects the cylinders’ radii which are critical to the lasing mode. We have shown that using a metal hard mask significantly minimizes the lateral reduction in the dimensions of the pattern in comparison to the HSQ hard mask. A metal mask allows for a precise fabrication of patterns defined by e-beam lithography. Moreover, we have discussed dry etch requirements to avoid etch pit formation in InP and the importance of an optimized wet etching window. A trapezoidal etching window with thin supporting arms was found to be the optimal window for a quick membrane release. We have shown that device functionality was unimpaired by either fabrication method. However, in the case of the metal hard mask, device performance such as a reduction in threshold power can be further improved with the reduction in sidewall roughness. Hence, a better alternative to PMMA spacer is needed: one that can be anisotropically etched straight with smooth sidewalls. The techniques outlined here will be of practical interest in the design and construction of novel membrane based devices with applications ranging from microfluidic biosensors to lasers on flexible substrates.