Experimental setup

TiN-Ti-caped Al interconnect lines approximately 0.5 µm thick were fabricated by a state-of-the-art 130 nm SiGe BiCMOS process on a 200 mm-diameter high-resistivity (10 kΩ·cm) Si wafer [Reference Mai and Kaynak5]. Lines with designed widths of 160 and 360 nm (actual widths 190 and 390 nm due to process bias as determined by SMM) were embedded in 1.3 µm thick SiO₂, then chemo-mechanically polished to a flatness of approximately 70 nm as indicated by the AFM profile in Fig. 1. Thereafter the wafer was diced into 25 mm × 15 mm chips. To simulate buried interconnect lines, some of the chips were coated with 20 nm of ALD Al₂O₃ at 300 °C. The conforming ALD did not alter the surface topography significantly (Fig. 1(b)), which minimizes the so-called “topography cross-talk” difference in SMM [Reference Smoliner, Huber, Hochleitner, Moertelmaier and Kienberger2] between samples with and without oxide coating. Otherwise, the SMM contrast between the two sets of samples would have been complicated by the difference in topography through its influence on the parasitic capacitance with the probe. Defining the image sharpness by the transition width from 25 to 75% height [Reference You, Ahn, Obeng and Kopanski3], Fig. 1(b) shows also that the AFM sharpness is approximately 80 nm, which is actually inferior to that of SMM, because the AFM scan parameters were not fully optimized giving the emphasis on SMM in this study.

Fig. 1. AFM (a) image and (b) linear scan of 190 nm-wide bare (—) aluminum lines. The image of buried lines is visually indistinguishable from that of bare lines and, hence, not shown. The profile of the buried lines is also very similar to that of bare lines as indicated by dashed line (- - -) in (b). Dashed line in (a) indicates where linear scans were taken.

The two sets of chips, with and without oxide coating, provided a good contrast between bare and buried metal lines. The aluminum lines on the uncoated chips were prevented from oxidation by TiN. The TiN and Ti layers, being nanometer thin, did not significantly change the electrical properties of the metal lines from that of pure Al. For tests in water, the chips were placed on the bottom of a petri dish that is filled with approximately 5 mm of deionized water, sufficient to immerse both the chips and the SMM probe, including its cantilever body for optical detection of the AFM deflection. The dielectric constant of deionized water was taken to be 70 according to the literature [Reference Ellison6]. The dielectric constants of Al₂O₃ and SiO₂ were measured to be approximately 7 and 4, respectively, from specially made metal-insulator-metal test structures by using an Agilent Technologies 4294A precision impedance analyzer. These dielectric constants would be valuable for in-depth data analysis.

Two chips, with bare and buried lines, respectively, were mounted side-by-side on the piezoelectric stage of a Keysight Technologies Series 7500 AFM/SPM system with a N9545C SMM nose cone. Typically, topography and microwave reflection coefficient (S ₁₁) were simultaneously measured during the same scan, and their changes plotted either in one-dimensional linear scans or two-dimensional images. Figure 2 shows the present SMM setup with the microwave signal channeled to a sharp probe for near-field interaction with the chip under test. The probe comprises a Rocky Mountain 12Pt400A full-metal platinum/iridium tip with 80 µm length and <20 nm radius, which was attached to a 400 µm-long, 60 µm-wide cantilever. With a spring constant of 0.3 N/m, the probe was sufficiently soft for direct contact with the chips. An Agilent Technologies N5230A PNA-L vector network analyzer (VNA) provided the microwave signal through an impedance matching network comprising a shunt resistor and coaxial cables. Typically, an input power of −5 dBm was generated at 5.371 GHz at which frequency the network was matched to the 50 Ω system impedance resulting in optimum signal-to-noise ratio (SNR). The magnitude and phase of S ₁₁ were measured by the VNA after passing through a 30 dB preamplifier. The scan size was typically 10 µm × 10 µm divided into 256 × 256 pixels. The SMM scan rate was limited to approximately 10 ms per pixel, because an intermediate frequency bandwidth of 1 kHz was used on the VNA for the optimum tradeoff between speed and SNR.

Fig. 2. (a) Schematic, (b) nose cone, and (c) stage of the present SMM setup.

Results and discussion

Figure 3 shows images formed simultaneously by AFM and SMM of 190 nm-wide bare aluminum lines. Whether an SMM image was formed by using the magnitude or phase of S ₁₁, it appears to have comparable contrast and sharpness to that of AFM. In reality, the SNR of SMM is not as good as that of AFM due to weak SMM signals. For example, in the present case, the contrast in S ₁₁ magnitude and phase are on the order of 0.01 dB (10 mdB) and 0.1°, respectively, comparable to that previously reported [Reference You, Ahn, Obeng and Kopanski3].

Fig. 3. Simultaneous (a) AFM, (b) SMM magnitude, (c) SMM phase images of 190 nm-wide bare aluminum lines.

Figures 4(a) and 4(b) show SMM magnitude images of 390 nm-wide bare and buried aluminum lines, respectively. It can be seen that, even when buried under 20 nm Al₂O₃, the lines are clearly discernable. Figure 4(c) shows that based on lateral scans across the middle of the image, bare and buried lines have comparable sharpness of approximately 70 nm. In fact, by averaging all linear scans across the lines, the sharpness was 69 ± 7 nm for bare lines and 70 ± 13 nm for buried lines. The present sharpness is comparable but not necessarily better than that of [Reference You, Ahn, Obeng and Kopanski3], because [Reference You, Ahn, Obeng and Kopanski3] buries the lines under a thicker oxide. Figure 4(c) shows also that the difference in contrast (signal magnitude) between buried lines and oxide is approximately half of that between bare lines and oxide. Had the oxide thickness not been known, it could have been estimated from the amount of the contrast degradation with the known dielectric constant of the oxide.

Fig. 4. SMM magnitude images of 390 nm-wide (a) bare and (b) buried aluminum lines. (c) Linear scans across both bare (—) and buried (- - -) lines.

Figure 4(c) shows that with a weaker signal, SNR is correspondingly worse for buried lines. SNR can be improved by averaging across adjacent linear scans, as will be illustrated for the scans in water. SNR may also be improved by averaging repeated scans or measurements (by VNA) at the expense of scan rate. Although presently the images formed by the S ₁₁ magnitude and phase are of comparable quality, this is not always the case. Depending on the detailed SMM working conditions, one can be significantly better than the other. For example, for different tips, samples, tip–sample distances, and ambient conditions, the resonant frequency can shift as will be illustrated for the case in water. The change of frequency can in turn change the loss and electrical length of the transmission line, and, hence, the magnitude and phase of S ₁₁.

As mentioned earlier, water was used to simulate high-k dielectric. Figure 5 shows the SMM phase image and linear scan over a 390 nm-wide bare line in water. It can be seen that although the line is still discernable, the signal and noise become comparable making it difficult to define sharpness from a single linear scan. However, the SNR can be significantly improved just by averaging a few adjacent scans as shown in Fig. 5(b). Statistically, a sharpness of 69 ± 13 nm was obtained, which is comparable to the cases without water. This is apparently an order-of-magnitude improvement over the 700 nm sharpness we reported in [Reference Jin, Xiong, Marstell, Strandwitz, Hwang, Farina, Göritz, Wietstruck and Kaynak4].

Fig. 5. SMM phase (a) image and (b) linear scans across an area of 3 µm × 3 µm (70 × 70 pixels) over a 390 nm-wide buried aluminum lines in water. In (b), averages across 6 (- - -) and 10 (…) adjacent lateral scans approximately 40 nm apart are shown for comparison with a typical single scan (―).

The sharpness improvement over [Reference Jin, Xiong, Marstell, Strandwitz, Hwang, Farina, Göritz, Wietstruck and Kaynak4] is mainly due to the difference in the parasitic capacitance between the probe body (cantilever) and the sample. Figure 6 illustrates schematically the capacitive interaction when a probe traverses an aluminum line embedded in silicon dioxide. In general, the capacitance can be divided into two parts, a sensing capacitance between the probe tip and the sample and a parasitic capacitance between the probe cantilever and the sample. Usually, with the tip in intimate contact with the sample and the cantilever suspended 80 µm in air above the sample, the parasitic capacitance is relatively small even though the cantilever is much larger than the tip. However, with the air replaced by water, the parasitic capacitance increases drastically. This parasitic capacitance not only overwhelms the sensing capacitance, but also remains constant so long as the cantilever completely overlaps a wide aluminum line such as in [Reference Jin, Xiong, Marstell, Strandwitz, Hwang, Farina, Göritz, Wietstruck and Kaynak4] (Fig. 6(b)). In this case, the line is effectively being probed by the large (400 µm × 60 µm) cantilever instead of the sharp tip, which explains why the sharpness is approximately 600 nm in [Reference Jin, Xiong, Marstell, Strandwitz, Hwang, Farina, Göritz, Wietstruck and Kaynak4], much larger than the tip radius of 20 nm. In comparison, when the probe traverses a narrow line (Fig. 6(a)), both the sensing and parasitic capacitances change readily, resulting in a sharpness of 70 nm, which is not much larger than the tip diameter. Thus, counterintuitively, in water it is more difficult to probe a large feature than a fine feature due the large parasitic capacitance with the cantilever. This parasitic capacitance can be minimized by more complete shielding of the microwave signal path all the way through the cantilever such as in a strip-line configuration [Reference Lai, Kundhikanjana, Kelly and Shen7].

Fig. 6. Schematic illustration of the capacitive interaction when an SMM probe traverses (a) narrow and (b) wide aluminum lines embedded in silicon dioxide.

For narrow lines, although water degrades the SNR, the sharpness remains approximately 70 nm as in air. This is very encouraging because it implies that SMM can also be used to non-invasively probe the internal structure of a live biological cell that needs to be kept in an aqueous solution as suggested by [Reference Tuca, Gramse, Kasper, Brinciotti, Oh, Campagnaro, Badino, Hinterdorfer and Kienberger8]. However, SMM of live biological samples presents additional challenges due to their soft and moving nature which need further development.

The present SMM image in water was obtained at 7.865 GHz instead of the typical 5.371 GHz. This is because the SMM uses an impedance matching network that works only at certain resonance frequencies. The resonance frequency shifted when the probe and the chip were immersed in water as shown in Fig. 7.

Fig. 7. Magnitude of reflection coefficient |S ₁₁| (after subtracting the 30 dB amplification factor) from an SMM probe in air (—) and water (….) without touching any sample.