Vertical and Smooth Single-Step Reactive Ion Etching Process for InP Membrane Waveguides

In this paper we present a novel single-step RIE process for InP membrane optical waveguide etching. The optimization of the process is focused on the sidewall verticality and surface roughness of the etched proﬁle. Signiﬁcant improvement on the etched proﬁle is achieved for the ﬁrst time in a single-step RIE process. Loss measurement on fabricated membrane waveguides etched with the proposed RIE process results in a record low waveguide propagation loss (2.5 dB/cm).

The InP membrane on Si (IMOS) 1,2 is regarded as the nextgeneration generic InP photonic integration platform. 3The intrinsic advantage of InP and its related compound material system allow for co-integration of a monolithic photonic layer with both active and passive functionalities on top of control electronics using an adhesive bonding technique. 4he dimensions of IMOS optical waveguides scale down dramatically as compared to traditional InP optical waveguides, 2 therefore the optical modes in the waveguide will experience higher losses due to much stronger interaction with the surroundings.A significant part of the optical loss comes from the roughness introduced during the processing of the membrane waveguide, especially from the lithography and the dry etching.The interaction between the waveguide surface (including sidewall and floor) and the fundamental TE mode in the membrane waveguide is shown in Fig. 1.It can be seen that both the roughness on sidewall and floor can strongly influence the optical loss.In our previous work, 5 the roughness induced during the electron beam lithography (EBL) step has been successfully reduced using a C 60 /ZEP mixed resist.A low propagation loss of 3.3 dB/cm for the fundamental transverse electric (TE) mode in the IMOS waveguide has been achieved in that work.Another source of roughness comes from the waveguide surface after reactive ion etching (RIE).It is known that the physical bombarding of reactive ions creates roughness during etching. 6The mechanism works mainly in two ways.One is the direct bombardment of reactive ions on the surface being etched.The other is the mask erosion with its induced roughness on the pattern edges, resulting in the transfer of this roughness to the sidewall of the waveguides.
The InP membrane is thermally isolated by a thick SiO 2 layer (see Fig. 1) during dry etching.The chip may heat up significantly if the ion bombardment is very intense.This can also cause roughness on the etched surface due to the change of polymer passivation condition.
Another effect of mask erosion due to ion bombardment is the selectivity of the etching (defined as the ratio of etched depth into the material to the consumed thickness of the etching mask).As currently the processing of IMOS devices utilize EBL at 20-30 kV electron acceleration voltage, 2,5 the thickness of the SiN x hard mask can influence the achievable feature size due to the backscattering of the electrons. 7A very thin (50 nm) layer of SiN x mask is currently used for the optimized EBL process, and this hard mask layer should be able to support about 1 μm deep etch into the membrane layers for the active device fabrication. 8erticality of the etched profile is also important for the IMOS waveguides with such high index contrast.Larger tilt of the waveguide sidewall will result in a stronger polarization crosstalk in waveguide bending regions. 9This can be detrimental for some of the photonic applications, for instance the arrayed waveguide gratings (AWGs) which z E-mail: y.jiao@tue.nlare sensitive to the polarization rotation in the waveguide arrays 10 and polarization-multiplexing coherent receivers where both polarizations carry signals. 11t is obvious that a dry etching process with low physical damage, high selectivity and good verticality is desired for the low-loss IMOS optical waveguide devices.The determination of the optimal process conditions is thus straightforward, mainly guided by concerns of roughness, selectivity and verticality.In this paper we present a single-step RIE process based on CH 4 /H 2 /Ar chemistry for the optimized etching of InP-based IMOS waveguides.Improved etched profiles together with smooth etched surfaces on InP membrane have been achieved with the proposed single-step process.
The rest of the paper is organized as follows.The motivation of using single-step RIE process and the determination of process parameters are discussed in Single-step RIE process section.In Verticality improvement and undercut removal section a novel technique on improving the sidewall verticality is analyzed in detail.Finally the characterization results on IMOS waveguides etched by the proposed process are presented in Waveguide measurement section, followed by conclusions.

Single-Step RIE Process
Two types of membrane samples are used during the etching.The cross-section geometries of the two sample types are depicted in Fig. 2. The thin membrane is used for passive waveguides and the thick membrane is used for amplifiers and detectors. 8The etch depths for thin and thick membrane samples were around 200 nm and 800 nm.
Etching is carried out in a SENTECH SI500 ICP machine operating at RIE mode (the ICP RF source being switched off).The determination of the process parameters depends on the requirements of the optical membrane waveguide application, as discussed in the introduction.The basic RIE process is chosen here due to its higher selectivity and smoother etched surfaces 12 as compared to alternatives (e.g.inductively coupled plasma RIE (ICP-RIE)).
The CH 4 /H 2 chemistry is preferred because of the low surface damage after etching, 13 the low consumption of mask material and good control of the etch depth.The major challenge for the CH 4 /H 2 based RIE process is polymer formation. 13The polymer formation on one hand can help to increase the selectivity by protecting the mask, but on the other hand can strongly influence the etched profile.Argon is added to the CH 4 /H 2 gas mixture to provide an extra degree of control on the polymer formation during etching. 14The ratio between methane and hydrogen is also important.A high hydrogen fraction in the gas mixture (e.g., the CH 4 /H 2 ratio is between 1 2 and 1 4 ) 15 can help to prevent spontaneous micro-masking 16 due to excessive polymer deposition.
Single-step etching scheme is preferred in the application of optical waveguides.Different from the cyclic process where CH 4 /H 2 etching x (µm) y (µm) is iterated with short O 2 descum to control the etched profile, the single-step process uses constant process condition throughout the entire etching.The advantage of the single-step process is that it significantly reduces process time and has the possibility of direct use of resist mask. 15So far the reported single-step processes for InPbased waveguide applications give a relatively poor sidewall angle (84-85 •15 ) as compared to the typical sidewall angle (> 88 • ) etched by cyclic processes iterated with oxygen plasma descum. 17The undercut beneath the hard mask due to the relatively large sidewall slope 15 in the reported single-step processes also limits the accuracy of the critical dimension definition in the waveguide devices.
The process pressure has a direct influence on the surface roughness after etching. 12Higher process pressure decreases the mean path of the reactive ions, which result in more collisions and more isotropic etching.Therefore surface roughness decreases as the process pressure increases.On the other hand high pressure also influences the verticality of the etched profile.Furthermore a high process pressure increases the etch rate due to the increase of the radical density.
The process parameters were determined with a series of tests defined in Table I.The tests are based on an original single-step RIE recipe from SENTECH.In this test, the thin membrane samples are The main focus of the tests was to study the influence of the CH 4 /H 2 gas ratio and the process pressure on the roughness of the etched surface.The platen temperature, RF power and argon flow were all kept constant during the tests.According to our observations, those three parameters have a negligible influence to the surface roughness.Moreover, the argon in the gas mixture helps to balance the polymer formation and consumption during the etching, and also helps the molecular pump to pump out lightweight H 2 more efficiently.After the etching, an O 2 plasma descum with 80 sccm O 2 flow, 75 mT process pressure, 150 W RF power is performed to remove the polymers on the sample and the sample holder.According to our observation, the polymers can be completely removed as long as the descum time is longer than 1/3 of the previous etching time.
Scanning electron microscopy (SEM) pictures of the etched thin membrane samples are shown in Fig. 3.It is clear from Fig. 3a-3c that the CH 4 /H 2 ratio has a significant influence on the surface roughness.This is mainly due to that fact that the polymer formation on the surface is dominated by the CH 4 fraction.A low CH 4 concentration in the chamber leads to less polymer formation and more sputtering.Furthermore it can be seen from Fig. 3c and 3d that a higher process pressure results in a smoother etched surface, due to the shift from physical sputtering toward chemical etching.
The waveguide cross-section profile etched with the bestperforming recipe (test 3) is also checked with SEM (see Fig. 4).In this test, both the thin and thick membranes are etched with the recipe from test 3.The cross-section of the etched waveguides confirms the smooth etched surface.However the relatively large sidewall slope and the resulting undercut beneath the SiN x mask is very significant.This observation matches with another recent report on single-step RIE process 15 in terms of sidewall slope and undercut.Moreover, the comparison between Fig. 4a and 4b shows that the  I.
undercut becomes larger as the etch depth increases.As has been discussed previously, both the large slope and the undercut have negative effects on the performance of the waveguide devices.Therefore an improved process with the same smoothness, but with more vertical sidewall and without undercut is desired.To our best knowledge, there are so far no reports on such a solution for single-step RIE processes.

Verticality Improvement and Undercut Removal
Here we report for the first time a single-step RIE process for InP waveguide etching with smooth surfaces, vertical sidewalls and no undercut beneath the hard mask.This is achieved by fine tuning the RF power level.A series of etching tests are carried out to examine the effect of the RF power on the etching results.The recipe from test 3 in the previous section is used, but now the RF power is scanned from 75 W to 250 W. Verticality, undercut, etch rate, selectivity and surface roughness are all investigated in detail.The verticality and undercut were determined from the waveguide cross-section pictures taken with the SEM.The etch rate is calculated from the step height measurements using a surface profiler.The selectivity is determined with step height measurements of the original SiN x mask thickness, the etched waveguide before and after removing the residual SiN x hard mask.Finally the roughness of the etched surface is measured using the atomic force microscopy (AFM) technique.
The detailed results are given in Fig. 5 to Fig. 7.It is clear from Fig. 5 that the verticality of the etched waveguide improves significantly as the RF power decreases.A sidewall angle of 88 • is obtained with an RF power of 100 W.This value is so far the best verticality achieved with a single-step RIE process for InP waveguide etching.The mechanism behind this can be the decrease of physical sputtering at low RF power.The reduction in RF power results in a decrease of plasma density and ion energy in the chamber.The reactive ions with low kinetic energy cannot sputter away the polymer passivation layer on the sidewalls.Therefore a vertical profile can be maintained.
On the other hand, the high-energy ions at high RF power can easily consume up the polymer passivation on the sidewalls and create large slopes and undercuts.The optimized sidewall angle can be obtained at a certain RF power level (100 W in our case) where the balance between polymer passivation rate and ion sputtering rate is achieved.It should also be noticed that the sidewall angle at 75 W RF power becomes slightly worse than that at 100 W.This can be due to the fact that the polymer formation on the sidewalls starts to dominate over the consumption.As the polymer passivation becomes thicker during etching, the effective mask width increases.The undercut beneath the hard mask is only observed for RF power levels of 200 W and 250 W. Below 200 W, the undercut is no longer visible in the SEM.
As can be seen from Fig. 6, the etch rate decreases almost linearly with the RF power.This is obviously due to the decrease of the plasma density at low RF powers.The selectivity between etched InP material and the consumed SiN x hard mask was always higher than 100, which is beneficial for the optical waveguide definition.The error bars for the selectivity are derived from the measurement uncertainty (0.1%) of each step height measurement (i.e., original SiN x thickness, waveguide heights before and after removing residual SiN mask erosion rate than the decrease of InP etch rate as the RF power decreases. The measured root mean square (RMS) roughness of the etched bottom surface is within the range of 0.6 -0.7 nm, as shown in Fig. 7.The obtained values are comparable to other literatures with reported low roughnesses. 12,15From the figure, there is no obvious indication on the influence of RF power to the surface roughness.However it is clear that the roughness increases slightly as the etch depth increases.The roughness on the etched sidewall is not measured with AFM due to its small sidewall dimension (less than 1 μm).The low sidewall roughness is verified by loss measurement in membrane waveguides which are fabricated with our proposed RIE process (see Waveguide measurement section).
According to the analysis above, the RF power of 100 W gives the best verticality, an acceptable etch rate, a high selectivity and a smooth surface.The optimized parameters for the process are summarized in Table II.The SEM pictures of etched waveguide structures on   II.
thin and thick membranes using this optimized process are shown in Fig. 8.The optimized recipe is used for the real membrane waveguide etching in the next section.

Waveguide Measurement
To further study the feasibility of the optimized process in realizing InP membrane structures, membrane waveguides with varying lengths are fabricated for propagation loss measurement.
The fabrication process of InP membrane waveguides is summarized in Fig. 9.It starts with the deposition of a 1850 nm thick SiO 2 layer on top of a Si wafer and a 50 nm SiO 2 layer on top of an InP wafer (containing the 300 nm InP waveguide layer) by using plasma enhanced chemical vapor deposition (PECVD).The InP wafer also contains a 300 nm thick InGaAs sacrificial layer between the membrane layer and the substrate.The Si wafer and the InP wafer are then adhesively bonded using BCB polymer [9] (see Fig. 9a).The thickness of the BCB layer after bonding is about 30 -50 nm.Next the InP substrate and the InGaAs sacrificial layer are wet-chemically removed, leaving the 300 nm thick InP membrane tightly bonded on top of the SiO 2 /Si carrier wafer (see Fig. 9b).The definition of membrane waveguides requires two steps of EBL.The first EBL (see Fig. 9c) starts with PECVD deposition of a 50 nm thick SiN x hard mask layer on top of the InP membrane.The C 60 -assisted EBL technique 5 is used in this fabrication to obtain the optimized resolution and sidewall smoothness on the resist pattern.This technique is based on a 100 nmthick ZEP resist material mixed with C 60 fullerene.The mixed resist is spin-coated on top of the SiN x layer, and the designed patterns are written on the resist layer using a Raith 150-two EBL system with 20 kV voltage and 10 μm aperture settings.After development, the resist patterns are transferred to the SiN x layer by means of CHF 3 RIE.Finally the patterns are transferred from the SiN x layer into the InP membrane using the optimized single-step RIE process from Table II.The first EBL step prints all the waveguide designs as well as the local markers for alignment of the second EBL step.The final InP waveguides have a width of 400 nm and an etch depth of 280 nm (see Fig. 9d).The waveguide is single-mode at our wavelength of interest (1550 nm).The second EBL step (see Fig. 9e) is similar to the first EBL, except that a pure ZEP resist with a thickness of 400 nm is used.This thickness results in a sufficient planarization of the etched step height from the first EBL and ensures a uniform surface for exposure.Grating couplers 18 for coupling light between optical fibers and the waveguides are then printed with a period of 680 nm, a filling factor of 50%, and an etch depth of 120 nm (see Fig. 9f).
An SEM picture of the fabricated waveguides is shown in Fig. 10.The lengths of the waveguides vary from 100 μm to 9 mm.There are two identical sets of waveguides on the fabricated chip to verify the results.
A commercial laser with 1550 nm wavelength and 6 dBm output power in fiber is used for the loss measurement.The laser light is coupled to the input grating coupler by means of a single-mode fiber   the record low achievement is mainly due to the advantage of the optimized single-step RIE process.

Conclusions
In this paper we reported a novel RIE process for etching InP optical membrane waveguides based on a single-step scheme and CH 4 /H 2 /Ar chemistry.For the first time it is shown that by fine tuning the RF power, the etched profile (i.e., verticality and undercut) can be significantly improved in the single-step RIE process.The optimized process provides almost vertical sidewall (88 • ), no undercut beneath the mask, high selectivity (130) and a smooth surface on the InP optical waveguides.Loss measurement on the fabricated membrane waveguides reveals a reduction of propagation loss using this optimized RIE process.Record low propagation loss on the InP membrane waveguides of 2.5 dB/cm has been achieved in this work.The reported process is a perfect tool for low-loss InP membrane waveguide etching.It also has huge potential in the etching of other types of InP-based optical waveguides.

Figure 1 .Figure 2 .
Figure 1.Schematic illustration of the interaction between the waveguide surface (including sidewall and floor) and the fundamental TE mode (electrical field intensity) in the membrane waveguide.

Figure 3 .
Figure 3. SEM pictures (angled top view) of the etched thin membrane samples using recipes from (a) test 1 (b) test 2 (c) test 3 (d) test 4 as listed in TableI.

Figure 5 .
Figure 5. Measured sidewall angle and undercut of the etched test structures as a function of RF power.

Figure 8 .
Figure 8. SEM pictures of etched waveguide structures on (a) thin and (b) thick membranes using the optimized process as listed in TableII.

Figure 11 .
Figure 11.Measured insertion loss of the InP membrane waveguides as a function of waveguide length.

Table I . Etching tests.
• C used with 50 nm-thick SiN x hard mask deposited on top of membrane.