Optically Monitoring and Controlling Nanoscale Topography During Semiconductor Etching

Supplementary Information

Optically monitoring and controlling nanoscale topography during semiconductor etching

Chris Edwards, Amir Arbabi, Gabriel Popescu and Lynford L Goddard

IMAGING CAMERA

The CCD is a Hamamatsu model ORCA-ER C4742-80 with 12-bit dynamic range and an 8.67 mm x 6.60 mm image sensor, 1344 x 1024 pixels, and a 6.45 μm x 6.45 μm pixel size. The focal lengths of lenses L3 and L4 are 7.5 cm and 40 cm, respectively, such that with a 10X objective, the overall transverse magnification is M = 53 and the field of view is approximately 160 mm x 120 mm.

ROTATING DIFFUSER

The surface of a circular piece of a polycarbonate was roughened by rubbing it against very fine grit sandpaper by hand. A direct current (DC) motor rotated the polycarbonate piece, which operated as the diffuser. Several pieces with different roughnesses were tested at various rotation speeds (on the order of 1000 rpm) and were characterized by measuring the spatial noise of the epi-DPM image with the diffuser in the light path. The combination with the minimum spatial noise was used for future experiments.

FREQUENCY DOMAIN NOISE ANALYSIS

To further understand the nature of the spatial and temporal noise and the effect of the diffuser, we computed the three-dimensional fast Fourier transform (FFT) of each series of 256 images of the flat sample and recorded the power spectral density (PSD) of the noise as seen in the sample plane. To keep the memory requirements to reasonable values, the images were cropped to retain the central 512 x 512 pixels. Figures S1a-b show the spatially averaged noise PSD. With the diffuser, the noise appears to consist of 1/f noise which dominates at low frequency and white noise which dominates at high frequency. Without the diffuser, the low frequency noise increases faster than 1/f as f approaches zero. The width of the noise spectrum is much narrower with the diffuser. This is because the diffuser randomizes the speckle pattern in time and enables the camera to average out the slowly varying speckle patterns. The baseline noise level at high frequency is slightly higher with the diffuser. This is probably due to the mechanical vibrations in the setup caused by the rotating DC motor. The small side peaks in Figure S1b may also be due to vibrational resonances in the setup. Figures S1c-d show the temporally averaged noise PSD. The noise is negligible outside the diffraction limited circle |k| = 2πNA/λ = 2.9 rad/μm,1 where NA is the numerical aperture of the objective. There is a very weak noise tail outside the diffraction limited circle along the kx = 0 and ky = 0 axes approximately 65 dB lower than the peak near DC. This noise tail, which breaks the rotational symmetry of the PSD, may be due to the grating alignment along the x direction and the nonlinear signal processing used to convert from phase to corrected height, which adds higher order components.

PROCESSING RECIPE FOR SPR 511A PHOTORESIST

Our recipe for SPR 511A photoresist (PR) begins with a standard degreasing followed by a dry bake at 125C for 4 minutes to remove any remaining moisture on the samples. HMDS:Xylene (1:2) is spun at 2000 rpm for 30 s, followed by the resist at 5000 rpm for 30 s to obtain a thickness of about 1 μm. A soft bake is then done at 90C for 90 s. The Karl Suss MJB-3 aligner C is used to do a 12 s exposure. Immediately after the exposure, a post-bake is done for 1 minute at 110C. The sample is then developed using MF-CD-26 for 25 s. A hard bake at 110C for 90 s is done before etching.

MICROPILLAR FABRICATION

The micropillars were fabricated using n+ GaAs wafers. The photolithography and developing were done using SPR 511A PR and the previously described recipe. A wet etch was then performed using a 1:1:50 solution of H3PO4:H2O2:H2O. Phosphoric acid is a standard wet etchant for GaAs. After the PR was removed, the samples were observed under the surface profiler and under the scanning electron microscope (SEM) to determine the height ranges for the given samples. Sample #1 was etched for 20 s and had heights of 60-70 nm. Sample #2 was etched for 40 s and had a height range of 120-130 nm. The heights measured by epi-DPM for both samples were within their respective ranges listed above.

MICROPILLAR IMAGE WITHOUT DIFFUSER

Figure S2 shows a micropillar from sample #2 imaged without the diffuser. The pillar is at a different location than the corresponding one of Figure 2c. Without the diffuser, the noise floor increased by 2.7 nm to 9.0 nm for the pillar and by 0.3 nm to 5.1 nm for the etched region. The mean height of this sample was 122.9 nm.

IN SITU MONITORING OF WET ETCHING

The sample containing the University of Illinois logo was also fabricated using n+ GaAs wafers. Plasma enhanced chemical vapor deposition (PECVD) was used to deposit 50 nm of silicon dioxide on the front and back sides of the GaAs sample. On the front side, SPR511A photoresist was spun and the logo was patterned using photolithography. The resist pattern was then transferred to the oxide using reactive ion etching (RIE) with Freon gases. The patterned oxide on the front side served as a hard mask for the wet etch process while the oxide covering the backside prevented depletion of the etchant from unwanted backside etching.

After patterning, the sample was placed upside down in a small Petri dish filled with water and propped up on two pieces of glass. The glass was used to allow the acid to diffuse under the inverted sample. The microscope was then adjusted until the sample was in focus on the eyepiece. The sample was then carefully adjusted until it was flat and then two drops of dilute acid (1:1:50 solution of H3PO4:H2O2:H20) were added to each side of the sample. The etch rate of PECVD silicon dioxide using such a dilute solution as 1:1:50 is less than 1 nm min-1. Given the short etch times, etching of the oxide can be considered negligible2.

A 3D movie (Video S1) was recorded showing the dynamics of the wet etching process. A total of 400 frames were collected at 8.93 frames s-1 for a total etch time of 44.8 s. The movie is replayed at real time speed for best representation of the etching dynamics. However, the slow changes can be more easily perceived when replayed at 2x speed. The color scale used in the movie is the same as that used in Figure 3a. At the beginning of the movie, before any etchant reaches the sample, the logo is slightly visible because there is a patterned oxide mask. It is not very visible because there is only a small index difference between the oxide and the water based ambient solution. At about 10 s, the etchant diffuses under the sample and begins etching the top left corner. The diagonal line in the lower left corner may be a scratch in the sample. At about 18 s, the etch depth in the narrow gaps between the lines of the logo, especially near the curl, is comparable to other unmasked portions; however near the end of the movie, one can see these gap regions do not etch as deeply. There are some artifacts in the movie. The tiny fluctuations in the tilt of the sample are due to the motion of the sample in the ambient etching solution. The quadratic fit process used to correct for sample tilt and phase curvature of the beam did not perform as well as on the flat sample because of the lack of large unetched areas in the image. The occasional sudden motions near the corners of the sample are attributed to phase unwrapping errors. Despite these minor errors, the general trends and dynamics of wet etching have been accurately captured in situ with epi-DPM.

CROSS-SECTIONAL PROFILES OF ETCHED STRUCTURES

Figure S3a shows a cross-section of the micropillar imaged with the diffuser from Figure 2c. The image was taken with a 10X objective (NA = 0.25) and has a field of view of 160 μm x 120 μm. The phosphoric acid wet etch on GaAs is isotropic. Thus, the lateral undercutting is approximately equal to the etch depth. The inset is taken from the edge of the pillar, represented by the dotted line, and shows that the sidewalls are vertical to about 0.5 μm. Thus, the edge resolution of our system is better than its 1 μm lateral resolution.

Figure S3b shows a similar cross-section of the University of Illinois logo. The image was taken with a 5X objective (NA = 0.1) and has a field of view of 320 μm x 240 μm.

CHARACTERIZATION OF MICROLENS ETCHING PROCESS

A spherical surface was fitted to the lens height data and the standard deviation of the residuals was calculated to quantify the performance of the lens. Figure S4a shows a cross section of the data, fit, and residuals. Figure S4b shows a top-down view of the residuals. The intended application is coupling arrays of vertical cavity surface emitting lasers (VCSELs) emitting at λ = 980 nm into 62.5 μm diameter multi-mode fibers. Thus, Figures S4b-c are limited to this diameter. Figure S4c shows a histogram of the residuals from S4b. The standard deviation of the residuals was 11.4 nm.

SUPPLEMENTARY REFERENCES

1 Saleh, B. E. A. & Teich, M. C. Fundamental of Photonics. 2nd edition edn, (John Wiley and Sons, Inc., 2007).

2 Williams, K. R., Gupta, K. & Wasilik, M. Etch rates for micromachining processes- part II. Journal of Microelectromechanical Systems 12, 761-778 (2003).