Atomic Force Microscopy on a chip

Article image: 
Michal Zawierta, Roger D. Jeffery, Gino Putrino, K. K. M. B. Dilusha Silva, Adrian Keating, Lorenzo Faraone and Mariusz Martyniuk
The University of Western Australia, Perth, WA 6009, Australia
Michal Zawierta was born in Poland in 1986. He received his BSc and MSc degrees in electronics and telecommunication from the Wroclaw University of Technology, Wroclaw, Poland, in 2011 and 2012, respectively, the B.Sc. degree in management from the Wroclaw University of Economics, Wroclaw, in 2011, and the Ph.D. degree in Electrical and Electronics Engineering from the University of Western Australia, Perth, WA, Australia. He is currently a Research Associate with the Microelectronics Research Group at the University of Western Australia, Perth, WA, Australia. His current research activities involve optical microelectromechanical systems.
We present an AFM probe with integrated on-chip optical interferometric readout based on silicon photonics. Our AFM probe combines the advantages of subnanometre resolution of optical readouts with on-chip miniaturisation. In the adopted approach cantilever deflection is determined using an integrated on-chip photonics waveguide by monitoring the separation between the sensing cantilever and an interrogating grating. The implemented methodology provides ultimate interferometric resolution and sensitivity, onchip miniaturisation and array scalability. This opens the door to a cost-effective AFM readout solution without sacrificing either measurement sensitivity, system miniaturisation, or the ability to be operated in multi-head configurations enabling fast large area AFM imaging.
We acknowledge the support from the Australian National Fabrication Facility (ANFF), Australian Microscopy & Microanalysis Research Facility, the Australian Research Council, Panorama Synergy Ltd., and WA State Government.
Corresponding Author
The University of Western Australia, M018, 35 Stirling Highway, Perth, WA 6009
Atomic Force Microscopy (AFM) is a force measurement and surface imaging technique which uses deflection of microcantilevers to scan surfaces of samples and provide useful information, such as surface topography. 
Most commercial AFMs use free-space optics to determine cantilever deflection via optical leveraging to translate small deflections of the cantilever tip into the significant motion of a laser beam on a quadrant detector. 
In this study, recently reported in the Journal of Ultramicroscopy14, the team at the University of Western Australia (UWA) made use of a silicon photonics circuit to create an integrated on-chip AFM probe with an optical interferometric readout. AFMs with external interferometric readouts have been studied in the past as an alternative to optical leveraging. However, such systems require precise on-going optical alignment, and so have been too challenging for commercial implementation. The approach presented here does not require any additional alignments as the microcantilever probe is intrinsically aligned with an interferometric readout during probe fabrication.
The fully on-chip miniaturised approach referenced throughout this paper as LumiMEMS™1–3 is schematically presented in Figure 1 and uses a silicon photonics circuit which contains an interrogating diffraction grating located under a cantilever. Laser light at 1550 nm is coupled to the chip via an optic fibre and guided within the silicon photonics waveguide to the interrogating grating. A proportion of the light is coupled out from the waveguide, reflected from the back-side of the sensing microcantilever and coupled back to the silicon photonics waveguide. The difference in the optical path length of the light that remained within the waveguide and the light reflected from the cantilever produces an interferometric effect. As the surface is scanned by the cantilever tip, the small change of the distance between the interrogating grating and cantilever produces a significant change in the output signal power proportional to the tip deflection and hence, surface topology.
Figure 1. (a) Artist impression of an isometric view of integrated on-chip cantilever deflection readout with the interrogating waveguide structure. The input light (PIN) is transmitted through the silicon photonics waveguide towards the interrogating grating, where part of the light is decoupled from the waveguide and directed towards the back side of the AFM microcantilever. The light reflected from the microcantilever is coupled back to the interrogating grating, modulating light transmission as a function of cantilever position. The modulated light (POUT) is guided through the waveguide to a detector. (b) Coloured scanning electron microscope (SEM) image of the fabricated device with marked buried silicon photonics waveguide.
This new approach has a range of advantages over the existing alternatives. First, the integration of readout with the sensing microcantilever allows for system miniaturisation. Integrated on-chip readout creates an extremely rigid structure and eliminates mechanical noise originating from optical leverage method used in most commercial systems.
Another advantage of this system is that by using commercially available silicon photonics chips and standard MEMS fabrication methods, the presented system can potentially provide a cost-effective AFM readout solution without sacrificing measurement sensitivity or increasing system complexity, providing a solution with an extremely low noise floor limited by photodetector shot noise. As demonstrated in this work, the system can be retrofitted into old AFM systems without significant changes in electronics or mechanical components, thus reducing readout noise and producing significant improvement in vertical resolution. 
Moreover, the system potentially allows sensing using multiple AFM cantilevers which could be used in imaging of large-area samples (e.g. full semiconductor wafers) with imaging speeds significantly surpassing those current commercial AFM systems can provide.
The scanning of the sample surface is schematically presented in Figure 1(a) and Figure 1(b) shows an SEM image of the fabricated device. The AFM probe was integrated into a commercial AFM system (Digital Instruments D3000 with DMLSG piezoelectrical scanner, manufactured in 1995) in an inverted configuration by placing the probe on the stage and having a calibrated test sample attached to the piezo scanner above the tip. During AFM imaging the interrogating light was provided by a 1550 nm single mode, low noise, laser source through SMF-28 optical fibres and a polarization controller since the grating couplers used in our interferometric readout are polarization sensitive. After being modulated by the cantilever motion the output power from the waveguide was coupled to an InGaAs photodetector (Thorlabs DET08CFC/M), amplified using transimpedance amplifier (Stanford Research Systems SR570) and interfaced to the DI D3000 as an input to the feedback loop signal using a Digital Instruments breakout box.
All AFM images presented in this paper were obtained in contact mode, with both open and closed feedback loop. Although open feedback is not commonly used in AFM imaging, imaging with an open feedback loop allows identification of the influence of noise from the Z-axis piezo-scanner on the experiment results. 
The AFM images were acquired using Digital Instrument D3000 software and then processed using the Gwyddion 2.48 software package4 using standard algorithms for data leveling by mean plane subtraction, alignment of the rows using the polynomial method, and shifting minimum (or average) data points to zero. The height distributions and RMS surface roughness were calculated using built-in statistical functions. For additional analysis, we used Fityk software5, which allowed all recorded probability density noise graphs to be fitted with Gaussian functions for calculation of peak parameters.
Electrostatic actuation of the cantilever was used to calibrate the cantilever position to the waveguide transmitted signal. This was performed using the fabricated on-chip electrodes, and the read-out response was characterised as a function of electrostatic actuation voltage and cantilever position measured simultaneously using a white light optical ZYGO NewView 7300 profilometer. During cantilever actuation, both the optical output power through the probe and the gap between the cantilever and substrate were measured independently. The electrostatic actuation and cantilever deflection noise density measurements were performed using a Femto DLPCA-200, ultra-low noise transimpedance amplifier with a bandwidth of 400 kHz in place of the Stanford Research Systems SR570 transimpedance amplifier that was used for imaging.
The signal spectrum measurements were recorded using an Agilent 89410A vector signal analyzer (VSA) based on the signal from the transimpedance amplifier and the AFM probe. Additionally, the Polytec OFV-5000 laser vibrometer was used to measure the cantilever displacement spectral response with a sensitivity of 50 nm/V. This measurement allowed confirmation of the value of resonant frequency and quality factor of the fabricated cantilever, as well as allowing verification of the sensitivity of the integrated readout. 
We commence the demonstration of the operation of the developed AFM LumiMEMS sensor module with the characterization of the cantilever motion in response to electrostatic actuation. Figure 2 presents the transmitted LumiMEMS signal as a function of cantilever downward deflection from its original rest position of 8.6 µm above the substrate. We observe a periodic response which confirms the interferometric effect taking place as the gap decreases with increasing actuation voltage. As expected, the peaks and nulls were found to repeat with a change in cavity distance between the cantilever and grating at approximately λ/(2 cos θ), where λ=1550 nm is the wavelength of the laser source6–7 and θ≅10° is the coupling angle of our grating8–9.
Figure 2. The optical power transmitted through the AFM module as a function of the downward deflection of the cantilever from its rest position of 8.6 µm above the substrate. The corresponding electrostatic actuation voltage is indicated on the top horizontal axis. Cantilever deflection was measured independently using an optical surface profiler. The two linear regions of positive and negative slope are marked on the plot.
The slopes between the consecutive minima and maxima are related to the minimum measurable change in the optical transmitted signal, δP, associated with the minimum resolvable change in the cantilever position, δz. Detailed knowledge of this transfer function was subsequently used to calibrate the probe for AFM measurements, and is directly related to the measurement sensitivity. As indicated in Figure 2, the positive slope before the first peak is
whereas the negative slope is
The full extent of both positive and negative slopes of the interferometric response can be used unambiguously. Our readout system provides up to 300 nm of linear dynamic range, which can be extended using two or more laser sources with different wavelengths6,10.
It is evident from Figure 2 that the peak transmission values for successive signal maxima increase as the cantilever downward deflection increases with increasing applied voltage. This is associated with the increasing optical cavity finesse for decreasing gap between the cantilever and the substrate. However, for the AFM measurements undertaken in this study, this change in peak transmission value is not relevant since all measurements were performed on the first positive slope, as indicated in Figure 2.
Acquisition of AFM images presenting a known surface topology of a reference sample was performed using the LumiMEMS readout and compared directly with a modern commercial AFM system (Bruker Dimension ICON). Figure 3 presents AFM images of the reference sample recorded using a commercial AFM tool (see Figure 3(a)) and using the LumiMEMS readout approach acquired in both open loop (see Figure 3(b)) and closed feedback loop (see Figure 3(c)) configurations. During measurements the LumiMEMS probe was integrated with the DI D3000 instrument, and the performance comparison is made with a Bruker Dimension ICON AFM instrument.
Figure 3. AFM images of a SiC-6H (0001) reference sample consisting of 0.75 nm high steps between consecutive terraces. The images cover an area of 2x2 µm2. (a) The image was acquired using a Bruker Dimension Icon AFM system with the settings set to achieve the best quality image. (b) The image was acquired using our integrated on-chip interferometric readout with an open feedback loop, which means that the piezo-scanner along the Z-axis remained rigid. (c) The image was acquired using our integrated on-chip interferometric readout with a closed feedback loop, which means that the piezo-scanner was modulated along the Z-axis to keep the scanning cantilever deflection constant. (d-f) The graphs show height distribution profiles for each AFM image.
For this experiment a silicon carbide reference sample was used: 6H-SiC (0001) supplied by TedPella, which has charasteristic steps of 0.75 nm in height fabricated on its top surface that correspond to half the lattice constant. The stated maximum average roughness of the terraces between steps is 90 pm. Figure 3(a-c) presents 2 x 2 µm2 surface area topography images consisting of 6 to 7 terraces with a vertical increment of 0.75 nm between the consecutive terraces. Figure 3(d-f) shows the probability density function (PDF) of the height distribution in the reference sample, such that each peak represents the level of a single terrace, starting from the lowest to the highest within the measured area.
The distances between the peaks represent the step heights, which can be used to verify linearity of the readout response. The amplitude and full width at half maximum (FWHM) of the peaks represent the sample surface roughness (noise) of each terrace. Narrower and higher peaks represent a lower value of surface roughness of the terraces, and Figure 3(d-f) lists the values obtained for the average distance between peaks and the FWHM for each measurement. The values for FWHM and RMS surface roughness are correlated by a factor of 2/√(2 in 2) for a Gaussian function11.
However, in our case, we did not take this approach as the RMS roughness calculated from FWHM will be an overestimate, since the FWHM is calculated for the whole image that also includes the steps. Figure 3(a-c) lists the average terrace surface roughness for all three images, noting that the surface roughness value obtained using the ICON instrument (62 pm RMS) is approximately double the value obtained using the LumiMEMS readout (37-38 pm RMS). The RMS roughness was calculated for all points located away from terrace edges as standard deviation of the recorded height values.
Subsequently, static non-scanning noise measurements were performed in order to determine the effective noise characteristic of the AFM probe installed in an AFM system. The measurement was conducted with the tip in contact with the sample at a fixed position, with the X-Y and Z-piezo scanners in static fixed positions, and measuring the signal under conditions typically used for surface imaging. This noise measurement was performed for the LumiMEMS readout attached to the D3000, and then repeated for the host D3000 optical beam deflection (OBD) readout, which allowed for a direct comparison of noise performance.
The results were logged as stationary AFM images presented in Figure 4, with the noise data logged line-by-line. Each line consists of 512 points recorded with a scan rate of 1 Hz, and each image consists of 512 lines. After the images were collected, the probability density functions (PDF) and RMS noise values were calculated. A significant difference in the noise PDF is evident between the standard Digital Instruments D3000 AFM setup and the LumiMEMS readout. We obtained an RMS noise value of 19 pm using the LumiMEMS readout, which was significantly lower than the 51 pm measured with the OBD readout of the Digital Instruments D3000. The reference noise measurements for the standard configuration of the Digital Instruments D3000 were repeated with three different commercial MikroMasch silicon cantilevers having spring constants of 0.18 N/m (HQ:CSC17), 2.8 N/m (HQ:NSC18) and 40 N/m (HQ:NSC15). There was no significant difference between the noise images and PDF measured with OBD readout for all three cantilevers. 
Figure 4. Probability density functions (PDF) and static AFM noise images for OBD and LumiMEMS readouts. The AFM images were acquired with a static AFM tip in contact with the sample, while the piezo-scanners were fixed and the images were created by recording the tip vibration over time. The PDF shows that the LumiMEMS readout is able to achieve a two-fold noise reduction in comparison to the OBD readout used in the Digital Instruments D3000 AFM system. The coloured Z-scale is applicable to both images. The full width at half maximum (FWHM) of the Gaussian noise distribution was used to verify calculations from the measurement noise values. The value of 19 pm RMS noise for LumiMEMS closely matches the noise figure obtained using the specifications of the Stanford Research Systems SR570 transimpedance amplifier of 18 pm RMS that was employed during image collection.
The reported measurements indicate that the LumiMEMS readout is able to achieve noise level magnitudes about half of that observed for the OBD readout. This statement is based on a direct comparison of the RMS noise values presented in Figure 4 for LumiMEMS (19 pm RMS) and the OBD (51 pm RMS), and the fact that this correlates very well with the doubling of surface roughness value obtained for the SiC terraces as measured by LumiMEMS (37–38 pm RMS) and OBD (62 pm RMS), as presented in Figure 3. In summary, by using the AFM system with the LumiMEMS readout, noise levels have been achieved that are well below those of commercially available AFM tools and that have been presented in previous reports in the literature. Our measurement of the static AFM noise of 19 pm RMS, outperforms the best values reported recently by Dukic et al.12, who achieved RMS noise levels of 25 pm for OBD and 32 pm for piezoresistive readouts.
The ultimate noise floor achievable by an AFM system utilizing our approach and operating in a common laboratory can be established by measuring the cantilever vibration as stimulated by the ambient laboratory environment. We characterise the deflection noise density (DND) of the cantilever13,14 by acquiring the frequency spectrum of the output optical signal transmitted through the LumiMEMS readout with the cantilever suspended above it. Figure 5 presents the DND of the cantilever vibration as a function of frequency measured using LumiMEMS, as well as for a simultaneous measurement performed using an external vibrometer. These measurements, as shown in Figure 5 indicate that Brownian motion dominates the measured frequency spectrum near the first resonant frequency of the cantilever, identified by the distinct peak near 45 kHz. Additionally, the second resonant frequency peak can be identified near 280 kHz in the LumiMEMS measurement and it is not visible in the vibrometer measured data since it is below the noise floor for this instrument. The Brownian motion of the cantilever, also known as thermomechanical motion, is not dependent on the readout technique and is a function of the cantilever design. As shown in Figure 5 and summarised in Table 1, the average effective measurement noise was found to be approximately DND=36 fm/√Hz and this level is well correlated to the limit estimated for our system due to utilized components and is dominated by the shot noise of the photodetector14
Figure 5. Deflection noise density spectrum for cantilever vibration stimulated by laboratory ambient. The green trace shows Brownian motion signal measured using an optical vibrometer and the pink trace is for the LumiMEMS interferometric readout. The effective measurement noise shown by the blue trace is dominated by shot noise, and is composed of dark current noise, photodetector shot noise, transimpedance amplifier electronics noise and noise of the laser source. The effective measurement noise was acquired using an optical attenuator fitted in place of the silicon photonics chip and set to give the same optical power at the receiver. The solid black line shows the square root sum of the squares of the calculated spectrum of the Brownian motion signal and the shot noise. The shot noise was estimated based on measured input power to the photodetector, Pt = 3.7 µW.
Table 1. Noise values for the characterised LumiMEMS readout compared to theoretical noise limits for the presented system configuration. The minimum detectable deflection (MDD) was calculated using the measured deflection noise density (DND) for a bandwidth (BW) of 87 kHz, which is the estimated signal bandwidth during AFM imaging.
We have demonstrated an AFM probe using an integrated on-chip interferometric cantilever deflection readout, which is based on silicon photonics, and combines the ultimate sensitivity of an interferometric optical readout with on-chip miniaturisation. The realised AFM probe, being characterised by a deflection noise density (DND) of 36 fm/√Hz, surpasses the performance of present day optical beam deflection (OBD) and piezoelectric readouts. Such a technology outperforms other readouts in terms of higher resolution imaging with potentially higher imaging rates. The integrated nature of this interferometric approach, since the AFM cantilever and readout are fabricated on the same chip, provides a system that requires no alignment of free-space optics. The presented LumiMEMS solution can potentially provide a cost-effective AFM readout solution without sacrificing measurement sensitivity, system miniaturisation or multiprobe array scalability.
The noise characteristics of the demonstrated sensing probe module were determined to be dominated by photodetector shot noise for frequencies well away from the cantilever natural mechanical resonance, where Brownian motion noise was found to dominate.
AFM images of reference samples have been presented using the developed probe, which demonstrated a significant reduction of the measured reference surface roughness compared to values obtained using a modern commercial AFM tool (Bruker Dimension ICON), achieving an AFM static image RMS noise floor of 19 pm.
1. J. Chaffey, “LumiMEMSTM : Sensor technology for emerging low power and IoT markets,” 28-May-2015. [Online]. Available: markets.html. [Accessed: 18-Nov-2017].
2. J. M. Dell, M. Martyniuk, A. J. Keating, G. M. Putrino, L. Faraone, and D. Silva, “System and method of performing atomic force measurements, U. S. Patent, US 2013/0205455 A1,” US20130205455 A1, 08-Aug-2013.
3. J. M. Dell, M. Martyniuk, A. J. Keating, G. M. Putrino, and L. Faraone, “Optical cantilever based analyte detection, U. S. Patent, US 2012/0218556 A1,” US20120218556 A1, 30-Aug-2012.
4. D. Nečas and P. Klapetek, “Gwyddion: an open-source software for SPM data analysis,” Open Phys., vol. 10, no. 1, pp. 181–188, 2011.
5. M. Wojdyr, “Fityk: a general-purpose peak fitting program,” J. Appl. Crystallogr., vol. 43, no. 5, pp. 1126–1128, Oct. 2010.
6. G. Putrino, M. Martyniuk, A. Keating, L. Faraone, and J. Dell, “On-chip read-out of picomechanical motion under ambient conditions,” Nanoscale, vol. 7, no. 5, pp. 1927–1933, Feb. 2015.
7. G. Putrino, A. Keating, M. Martyniuk, L. Faraone, and J. Dell, “Model and Analysis of a High Sensitivity Resonant Optical Read-Out Approach Suitable for Cantilever Sensor Arrays,” J. Light. Technol., vol. 30, no. 12, pp. 1863–1868, Jun. 2012.
8. D. Taillaert et al., “Grating Couplers for Coupling between Optical Fibers and Nanophotonic Waveguides,” Jpn. J. Appl. Phys., vol. 45, no. 8R, p. 6071, Aug. 2006.
9. L. Carroll et al., “Photonic Packaging: Transforming Silicon Photonic Integrated Circuits into Photonic Devices,” Appl. Sci., vol. 6, no. 12, p. 426, Dec. 2016.
10. G. Putrino, A. Keating, M. Martyniuk, L. Faraone, and J. Dell, “Integrated Resonant Optical Readout Applicable to Large Arrays of MEMS Beams,” IEEE Photonics Technol. Lett., vol. 24, no. 24, pp. 2243–2246, Dec. 2012.
11. E. W. Weisstein, “Full Width at Half Maximum.” [Online]. Available: [Accessed: 19-Apr-2018].
12. M. Dukic, J. D. Adams, and G. E. Fantner, “Piezoresistive AFM cantilevers surpassing standard optical beam deflection in low noise topography imaging,” Sci. Rep., vol. 5, p. 16393, Nov. 2015.
13. K. Zinoviev, C. Dominguez, J. A. Plaza, V. J. C. Busto, and L. M. Lechuga, “A novel optical waveguide microcantilever sensor for the detection of nanomechanical forces,” J. Light. Technol., vol. 24, no. 5, pp. 2132–2138, May 2006.
14. M. Zawierta et al., “Atomic force microscopy with integrated on-chip interferometric readout,” Ultramicroscopy, vol. 205, pp. 75–83, Oct. 2019.
Website developed by S8080 Digital Media