Active mechanical noise cancellation scanning tunneling microscope

REVIEW OF SCIENTIFIC INSTRUMENTS 78, 073705 2007 Active mechanical noise cancellation scanning tunneling microscope H. Liu, Y. Meng, H. W. Zhao, and D. M. Chen a Beijing National Laboratory for Condensed Matter Physics, Institute of Physics, Chinese Academy of Sciences, Beijing 100080, China Received 7 February 2007; accepted 13 June 2007; published online 6 July 2007 We present the design and performance of an active mechanical noise cancellation scanning tunneling microscope STM. This system features two key parts: a twin-tip scanner and an active mechanical noise cancellation algorithm. The twin-tip scanner functions as two independent STMs which share nearly the same mechanical transfer function, allowing both STMs to sense nearly identical background mechanical noise. Based on an adaptive digital signal processing technique, the active mechanical noise cancellation algorithm applies the noise sensed by the first STM to concurrently cancel the noise in the second STM and hence allows the second STM to acquire spectroscopy with a significantly improved signal to noise ratio. This system demonstrates long-term stability of the tip-sample tunnel junction and improved spectroscopy measurement in a mechanically noisy environment. 2007 American Institute of Physics. DOI: 10.1063/1.2755399 I. INTRODUCTION a Electronic mail: dmchen@aphy.iphy.ac.cn Since the invention of scanning tunneling microscope STM, 1 tunneling spectroscopy has proven to be a valuable tool for investigating the electronic properties of a wide range of materials at atomic scale owing to its unique ability to directly access the local density of states of the specimen. The spectroscopy measurement usually requires long-term stability of the tip-sample tunnel junction and effective isolation of the STM mechanical components from ambient noise such as mechanical and acoustical vibrations is essential. With this in mind, a great deal of effort has been devoted to improve the performance of spectroscopy by designing a compact positioning device 2 4 and isolating the STM head from external mechanical and acoustical noise sources. Common mechanical and acoustical noise isolation techniques are passive, such as spring suspension and magnetic damping. 5 7 However, there has been little effort on active mechanical noise control to improve the signal to noise S/N ratio of the tunneling spectroscopy. Interestingly, the extremely high sensitivity of STM to the tip-sample separation makes it a superior mechanical noise sensor. Would it be feasible then to use one STM as the sensor to actively sense the mechanical noise and use this to cancel the same noise sensed by a second STM dedicated to spectroscopy? Here we report our development of an active mechanical noise cancellation STM system and demonstrate how such active mechanical noise sensing and cancellation is realized. In the following sections, we first outline the system architecture and describe the design of the two identical STMs and their mechanical characteristics. Next, the details of the active mechanical noise cancellation algorithm are introduced and the simulation result is presented. Finally, the performance of the active mechanical noise cancellation STM is evaluated and analyzed through the stability of tipsample tunnel junction and spectroscopy taken on Si 111-7 7 in mechanically noisy environment. II. INSTRUMENT DESIGN A. System architecture Figure 1 outlines the basic architecture of this new technique. The system comprises of two STMs each having its own servo control unit. During the spectroscopy measurement, one of the two STMs runs in closed-loop feedback mode and plays the role of a mechanical noise sensor, whose signal, after proper processing, is fed forward into the second STM running in open-loop feedback mode. This results in active cancellation of the mechanical noise in the second STM while taking the spectroscopy. While there are a number of ways to implement the mechanical design, we chose to assemble two identical tip holders into one common scanner, where the scanner is fixed on an X stage directly attached to the bottom of a helium Dewar. The STM Dewar is a part of an ultrahigh vacuum UHV system and connected to another chamber equipped with standard surface preparation and analysis apparatus. The STMs are designed to function in a low temperature and UHV environment. To limit the mechanical shocks introduced by the floor vibration to the STM head, a set of air legs is used to suspend the entire UHV system. To further reduce the coupling of acoustic noise to the STM, the system is installed in an acoustically shielded room. However, no additional springs, O-rings, or other types of vibration isolation schemes are inserted between the STM and the Dewar. This allows rapid cooling of the sample in a low temperature experiment. The active mechanical noise cancellation is implemented in a digital signal processor DSP to achieve real time learning of the system mechanical noise and tuning of the computation model. 0034-6748/2007/78 7 /073705/5/$23.00 78, 073705-1 2007 American Institute of Physics

073705-2 Liu et al. Rev. Sci. Instrum. 78, 073705 2007 FIG. 1. Color online The basic architecture of the active mechanical noise cancellation technique. B. Twin-tip scanner The design goal of the two-stm mechanical system is to yield as closely matched mechanical response as possible for the two STMs. Therefore, we chose to have the two STMs share the same scanning, positioning, and other supporting structures. Figure 2 a shows the twin-tip scanner consisted of three parts: the XY piezoelectric scanning segment, the Z piezoelectric segment, and the two identical tip holders. The XY segment and the Z segment are fabricated from a single piece of piezoelectric tube which is fixed to an X-positioning stage. The X-positioning stage is a piezoelectric driven stepper that translates both tips parallel to the specimen surface. The Z piezoelectric segment is separated into two halves, each for one STM s Z motion. There are two outer electrodes, Z 1 and Z 2, for the two tips, respectively, and one inner electrode shared by both tips. The Z piezoelectric segment serves as a driver for the tip forward/backward actuation as well as for the inertial motion. As shown in Fig. 2 b, each tip holder is supported by four sapphire balls glued at the inner surface of the Z piezoelectric tube and is held against the balls by a CoSm magnet glued at the inner surface of the Z piezoelectric tube. Both tip holders are carefully fabricated to share nearly the same mass and shape in order to achieve FIG. 2. Color online a The mechanical model of a twin-tip scanner with a diameter of 8 mm. It consists of an XY scanning segment and a Z segment. The Z segment is splitted into Z 1 and Z 2, each supporting a tip holder. b Front view: the inner design of the twin-tip scanner. Each tip holder is supported by four sapphire balls and held by a CoSm magnet. The distance between these two tips is about 4 mm. FIG. 3. Mechanical transfer functions as a function of frequency from 1Hzto1kHz. a The amplitude transfer property and b the phase transfer property. Both STMs share the same resonance frequency f c =515 Hz. matching mechanical response. A 100 m diameter soft gold wire is used to connect the tip to a nearby fixed electrode so not to affect the movement of the tip holder. The forward/ backward motion of the tip to/from the specimen surface is realized by applying a DSP generated well-known slip-stick waveform to the outer electrode of Z 1 or Z 2 piezoelectric segment. For the fine approach, Z 0 is used to prevent the tip from crashing into the sample. During the stepping, Z 0 is biased to fully retract the tip, and between the steps Z 0 is biased to extend the tip toward the surface to check for possible tunneling current. A feedback is turned on to hold the tip in place as soon as a tunneling current is sensed. The mechanical responses of the two STMs are characterized by measuring their mechanical transfer function, as shown in Fig. 3. We sweep the frequency of a 10 mv peakto-peak sine wave form drive signal applied to Z 0 from 1 Hz to 1 khz while recording the respective amplitude and phase responses of Z 1 and Z 2 in closed-feedback mode. To avoid phase change while ramping the frequency, we varied the frequency with a slow rate of 1 Hz/s. The amplitude of the drive wave form is set to prevent the tips from crashing into the surface. From the results of Fig. 3, we conclude that STM-1 and STM-2 share nearly the same resonant frequency as shown in the amplitude plot Fig. 3 a and phase plot Fig. 3 b. The amplitudes for both STMs are nearly the same and their phases also coincide and increase smoothly with increasing frequency, except at the resonant frequency of the STM, which is low compared to that of the piezotube. 8,9 We thus conclude that our two STMs mechanical performances are well matched and should respond identically both in amplitude and in phase to the ambient mechanical noise up to the STM s lowest resonant frequency 515 Hz, as shown in Fig. 3. C. Control system Figure 4 a shows the schematic diagram of the STM control system which consists of three parts: a host personnel computer PC, a DSP, and homemade front-end electronics including analog-digital converter A/D, digital-analog converter D/A, high voltage HV amplifier, and preamplifier. The PC is responsible for sending user commands to and receiving data from the DSP through a serial port interface.

073705-3 Active mechanical noise cancellation STM Rev. Sci. Instrum. 78, 073705 2007 transducer gain at different temperatures, especially at room temperature RT, liquid nitrogen temperature, and liquid helium temperature. The preamplifier is designed to have a gain of 1 V/1 na with a 2.5 khz bandwidth. FIG. 4. Color online a Schematic diagram of the STM servo control system implemented in a DSP and interfaced via the front-end electronics including A/D, D/A, preamplifier, and HV amplifier to control the STM. b Schematic diagram showing three modes of operation of the STM control system. c The simulated performance of the adaptive FIR filter. Due to the limited bandwidth of our current DSP, 10 the STM XY scan function is provided by other homemade scan drivers that are controlled and synchronized by the PC. The primary function of the DSP is to control the three Z electrodes to accomplish various functions such as tip stepper motion, the two independent STM servos, and active mechanical noise cancellation. The sampling rate of A/D is 10 khz, which is sufficient for normal STM feedback. The STM feedback is realized by an interrupt service running in the DSP. Three D/A outputs are used to drive the Z 1, Z 2, and Z 0 via HV amplifiers, and two A/D inputs are used to acquire the tunneling currents I 1 and I 2 after current-to-voltage conversion in the preamplifiers. Each HV amplifier has four gains to achieve high resolution topography and proper D. Active mechanical noise cancellation algorithm Figure 4 b shows the principle of the feedback and active mechanical noise cancellation algorithms running in the DSP. It has three modes of operation: normal, learning, and active, as shown in the table see Fig. 4 b. In normal mode, both STMs are controlled by an incremental proportional integrator PI. 11 The incremental PI adequately maintains good loop stability and routinely achieves atomic resolution in the imaging mode. 12 The learning and active modes adapt different feedback schemes, as will be discussed in detail below. In our design, both STMs are functionally identical and their respective roles can be swapped when instructed by the user. Figure 4 b illustrates the case where STM-1 is used for the spectroscopy measurement and STM-2 is used as a mechanical noise sensor, and we will keep this assignment for the remaining of this article. In normal mode, the spectroscopy measurement is performed with open-loop feedback, and it will become unstable or noisy if the ambient mechanical noise exceeds the acceptable level. In active mode, during the spectroscopy measurement we feed forward dz 2 to dz 1 and STM-1 is effectively running with closed-loop feedback though it is set to run in open-loop feedback mode. Although the two STMs mechanical responses are closely matched, the driving signals dz 1 and dz 2 are not always identical due to other factors such as the slightly different amplitude and phase shift caused by their relative position with respect to the sample. Occasionally, the amplitude of the fed-forward signal dz 2 varies with time and is difficult to track manually. This can lead to artificial features in the measured spectra. To overcome this problem, we use an adaptive finite impulse response FIR filter 13 to process dz 2 before it is fed forward into dz 1. As shown in Fig. 4 b, the adaptive FIR filter is to generate the dz 1 signal from dz 2. The coefficients of the FIR filter are updated only in learning mode using the least mean square LMS method 14 and the error is computed as e n =dz 1 n dz 1 n, which measures the difference at the nth iteration between the output of the adaptive filter and the STM-1 driving signal dz 1 n in closed-loop feedback mode. Based on the calculation, the adaptive filter adjusts its coefficients to reduce the difference and hence optimizes the fedforward signal dz 1 from the FIR filter. The coefficients are given by h n+1 i = h n i + e n dz 2 n i, 1 where h i is the ith coefficient of the adaptive FIR filter and is the learning step which controls how quickly the adaptive filter converts dz 2 to dz 1. When is very small, the coefficients change only a small amount at each iteration, and the filter converges slowly. With a large more gradient information is included in each iteration, and the filter converges more quickly. However, when is too large, the coefficients may change too quickly and the filter will diverge or oscillate.

073705-4 Liu et al. Rev. Sci. Instrum. 78, 073705 2007 In order to demonstrate the performance of the adaptive FIR filter, we did a simulation in MATLAB Ref. 15 using an input dz 2 =A 2 sin 2 ft+ to generate dz 1 =A 1 sin 2 ft, as shown in the upper panel of Fig. 4 c. During the learning phase, the output signal dz 1 solid line from the adaptive filter tracks the signal dz 1 while updating FIR filter coefficients. During the active phase, the signal dz 1 completely matches the signal dz 1. The adaptive filter works well for all ranges of from to + and even for a large amplitude difference. In general, the spectrum of the adaptive filter input signal, such as dz 2, has to cover the spectrum of signals to be simulated, such as dz 1, in order to give accurate predictions. For the active mechanical noise cancellation application, the adaptive filter coefficients have to be updated periodically in order to effectively simulate the ambient mechanical noise which may change in a long duration. If the ambient mechanical noise varies too quickly, the adaptive filter coefficients need to be updated in every learning cycle and cannot be used to generate the reference signal in the active mode. Thus, the active mechanical noise cancellation fails in this case. In order to inform the user whether it is the time to take spectroscopy measurement, we defined an indicator as Q 2 = i h n i h n 1 i 2 to give the stability of the adaptive filter. If Q 2 goes below a predefined level within a reasonably long time, like 10 s, it indicates that the adaptive filter is stable and can accurately generate the reference signal. As shown in the lower panel of Fig. 4 c, after several cycles of learning, the Q 2 goes below the defined level and does not change in subsequent learning steps. In this case, we have the accurate prediction of dz 1 from dz 2. III. SYSTEM PERFORMANCE To test the performance of the above described active mechanical noise cancellation technique in a systematic way, we generate a noisy environment using a speaker with a variable frequency source to generate acoustic vibration in the Z direction of the twin-tip scanner. The results show that the system maintains a good stability of the tip-sample tunnel junction for an extended period of time in the open-loop feedback mode, and the S/N of spectroscopy taken on the Si 7 7 is significantly improved for a mechanically noisy environment. We detail these tests and discuss the results below. A. Tip-sample tunneling junction stability in active mode 1. Response to a controlled mechanical excitation In order to systematically evaluate the frequency dependence of the tip-sample tunneling junction stability, we sweep a speaker s input frequency from 1 Hz to 1 khz and observe the tunneling current I 1 while the STM is running in all three modes defined above. The amplitude of the speaker input signal is adjusted so that the level of coupled vibration at the tip is still correctable by the servo. Figure 5 a shows the I 1, dz 1, and dz 2 as a function of time in various modes with a 10 Hz simulated noise. As described in Sec. II D, STM-2 works as the noise sensor and has to run in normal FIG. 5. a dz 1, dz 2, and I 1 acquired in three modes as indicated under the controlled 10 Hz mechanical noise. b Comparison of the Fourier transform of I 1 between the active mode lower curve and the normal open-loop feedback mode upper curve with offset under the ambient mechanical noise. The inset is the corresponding tunneling current I 1 as a function of time. c Numerical di/dv spectra acquired in the normal mode offset and in the active mode under the ambient mechanical noise left panel and a controlled 10 Hz mechanical noise right panel. closed-loop feedback mode all the time, and STM-1 runs in one of the three modes. In learning mode, STM-1 runs in closed-loop feedback and the adaptive FIR filter coefficients are updated by comparison of dz 1 and dz 2 signals. When STM-1 is switched to the active mode with open-loop feedback following the learning phase, the current I 1 shows good stability and the noise level is comparable to that in the closed-loop feedback mode, indicating that the active mechanical noise cancellation is effective. When the active mechanical noise cancellation is put in hold when STM-1 remains in open-loop feedback, the current I 1 exhibits a dramatic increase of a 10 Hz noise correlated to the acoustic vibration source. From this test, we conclude that the active mechanical noise cancellation technique can indeed be used

073705-5 Active mechanical noise cancellation STM Rev. Sci. Instrum. 78, 073705 2007 for improving the S/N of the tunneling spectroscopy. In general, this technique works well except when the noise frequency is around the resonant frequencies of the STM. Thus, noise reduction is most effective when the noise frequency is lower than the lowest resonant frequency of the STM. If there is no shock noise, the active mechanical noise cancellation can maintain a stable tunnel junction for 10 20 s which is sufficient for a spectroscopy measurement. To optimize the performance, typically a new learning is needed every 20 s as the relative response of the two STMs to the ambient mechanical noise changes. 2. Response to the ambient mechanical noise Next, we examine the tip-sample tunneling junction stability under the ambient mechanical noise. In our laboratory, a common mechanical noise comes from the mechanical vibration of a nearby power transformer which results in a 50 Hz noise in the tunneling current. Figure 5 b shows the comparison of the tip-sample tunneling junction stability between the active and normal open-loop feedback modes. The frequency spectra show that the active noise cancellation suppresses more than 90% of the main noise 50 Hz of a normal open-loop feedback signal. The inset of Fig. 5 b shows a good stability of the tunneling current in active mode for an extended period of time. It is obvious that the active mechanical noise cancellation can also be used to compensate for the thermal drift of the STMs. B. Spectroscopy performance The quality of the spectroscopy in active mode is tested using a Si 111-7 7 specimen. Figure. 5 c compares the spectroscopy numerical di/dv acquired in normal openloop feedback mode with one taken in active mode with open-loop feedback in the ambient mechanical noise left as well as in a 10 Hz noisy environment generated by an acoustic source right. During the data acquisition, each measurement point takes 30 ms to yield good signal averaging. It is evident that in both cases, especially with the 10 Hz noise, the spectra measured in active mode shows considerably less noise. Thus, the active mechanical noise cancellation technique is effective in the control of mechanical noise of the STM and hence improves its S/N during spectroscopy measurement. C. Limitation of the system The principle of the active mechanical noise cancellation STM requires that both STMs sense the same mechanical noise spectrum. In practice, however, there will always be some minor differences in the mechanical responses of the two STMs, especially in higher frequencies. Therefore, the unmatched noises could become an artifact in the measurement. Moreover, noise appeared in the tunneling current that is originated from sources other than mechanical or acoustical noise can also lead to an artifact. Capacitive coupling due to the change of tip bias is one such example. Thus, during the spectroscopy measurement, one has to vary the gap voltage with a small incremental step usually 10 mv to avoid coupling of dv 1 into I 2. IV. SUMMARY We have presented the design and performance of an active mechanical noise cancellation scanning tunneling microscope aimed for improving the performance of tunneling spectroscopy in ambient condition. We have shown that a novel twin-tip scanner design can, indeed, yield two STMs with nearly the same mechanical transfer function. The concept of using one STM as a sensor to feed forward its signal to cancel the noise of the second STM works well. An adaptive digital signal processing and learning algorithm enables the automation of an intelligent control for the active mechanical noise cancellation without constant user intervention. This new type of STM will be quite useful for fine spectroscopy measurement with simple system mechanical isolation. ACKNOWLEDGMENTS The authors thank Sergy Prydkin for his technical contribution to the project. The financial support of this project from the National Science Foundation of China Grant No. 90406017 and from the Chinese Academy of Sciences is gratefully acknowledged. 1 G. Binning and H. Rohrer, Helv. Phys. Acta 55, 726 1982. 2 S. H. Pan, E. W. Hudson, and J. C. Davis, Rev. Sci. Instrum. 70, 1459 1999. 3 C. Dubois, P. E. Bisson, A. A. Manuel, Ø. Fischer, and S. Reymond, Rev. Sci. Instrum. 77, 043712 2006. 4 J. Wiebe, A. Wachowiak, F. Meier, and D. Haude, T. Foster, M. Morgenstern, and R. Wiesendanger, Rev. Sci. Instrum. 75, 4871 2006. 5 Omicron Nanotechnology GmbH, D-65232 Taunusstein, Germany. 6 Unisoku Co., Ltd, Osaka 573-0131, Japan. 7 B. Koslowski, Ch. Dietrich, A. Tschetschetkin, and P. Ziemann, Rev. Sci. Instrum. 77, 063707 2006. 8 D. Croft and S. Devasia, Rev. Sci. Instrum. 70, 4600 1999. 9 E. Anguiano, A. I. Oliva, and M. Aguilar, Rev. Sci. Instrum. 69, 3867 1998. 10 ADSP-21061 is a DSP product of Analog Devices, Inc. 11 A. O Dwyer, Handbooks of PL and PID Controller Tuning Rules Imperial College Press, London, 2003. 12 The digital PI feedback gives the atomic resolution image of Si 111-7 7 and image of Pb island grown on Si 111 substrate. 13 S. Haykin, Adaptive Filter Theory, 3rd ed. Prentice-Hall, Englewood Cliffs, NJ, 1996. 14 S. Haykin and B. Widrow, Least-Mean-Square Adaptive Filters Wiley, Hoboken, New Jersey, 2003. 15 MATLAB is a technical computing platform of The MathWorks, Inc.