Design and Material Contributions to Second-Harmonic Nonlinearities in RF Silicon Integrated Passive Devices

Transcription

1 Design and Material ontributions to Second-Harmonic Nonlinearities in RF Silicon Integrated Passive Devices by Robert Frye, RF Design onsulting, LL; Robert Melville, Emecon, LL; and Kai Liu, STTS hipp, Inc. opyright Reprinted from 2014 Electronic omponents and Technology onference (ET) Proceedings. The material is posted here by permission of the IEEE. Such permission of the IEEE does not in any way imply IEEE endorsement of any STTS hipp Ltd s products or services. Internal or personal use of this material is permitted, however, permission to reprint/republish this material for advertising or promotional purposes or for creating new collective works for resale or distribution must be obtained from the IEEE by writing to pubs-permission@ieee.org. y choosing to view this document, you agree to all provisions of the copyright laws protecting it.

2 Design and Material ontributions to Second-Harmonic Nonlinearities in RF Silicon Integrated Passive Devices Robert Frye, Robert Melville* and Kai Liu** RF Design onsulting, LL 334 arlton venue Piscataway, NJ US * Emecon, LL 1756F Springfield ve. New Providence, NJ bstract Second harmonic distortion in silicon integrated passive devices is sufficiently low to be of no concern in most applications. However, recent changes in frequency allocation for 4G cellular applications have raised concerns for interference with sensitive GPS signals. Previously, we reported a number of measurements on low-pass filters around 750MHz. These measurements suggested that harmonic distortion was mainly attributable to lateral surface fields in the circuit s devices interacting with the substrate depletion layer. If this is the case, the nonlinearities in some cases could be reduced by design changes. We have performed further experiments and measurements aimed at improving our understanding of these issues and the relative importance of design versus substrate material. We have evaluated low pass filters with design variations aimed at reducing the interaction of the filters with the substrate. The designs were fabricated on substrates of different nominal resistivity, and with and without surface treatments aimed at reducing the so-called parasitic surface conduction effect. In substrate wafers that do not have surface passivation, harmonic distortion is generally high. In this case, we find that minor design variations may significantly reduce the second harmonic power. Furthermore, we observe in these devices that the harmonic power increases under conditions of illumination, suggesting that surface conduction plays a role. Importantly, however, in surface-passivated wafers we find that the harmonic power is significantly reduced and that design variations make less difference. Furthermore, we observe insensitivity to illumination, indicating that the photoconductive effect in these passivated samples is negligible. We have compared samples fabricated on different substrate resistivities. We find a very strong dependence. omparison of samples fabricated on silicon substrates with 3500Ωcm resistivity, versus 1000Ωcm, shows roughly an order of magnitude reduction in second harmonic power. For 4G cellular applications in the 700MHz frequency band, substrate resistivity appears to be a critical consideration. s in other studies, we find that 2nd-harmonic generation is closely correlated with losses in passive components and circuits. We have examined inductor quality factor, as well as insertion losses in transmission lines and filters, and find that higher losses are generally associated with higher harmonic power levels /14/$ IEEE ** STTS hipp, Inc West Greentree, Suite 117 Tempe, rizona 85284, US kai.liu@statschippac.com Introduction To increase the available bandwidth for 4G cellular telecommunications, the F in 2008 made available new spectrum in the range from 668 to 806MHz (the so-called 700MHz band). Signal propagation in this band is relatively good at penetrating walls, making it especially desirable for cellular applications in dense, urban areas. drawback, however, is that for signals near the middle of this band, second harmonic power falls within the sensitive Global Positioning Satellite (GPS) frequency of 1575MHz, as shown in Fig. 1. To help avoid interference, within the 700MHz the sub-band around 788MHz is not used for cellular communications, and serves as a guard band. Guard band MHz 2X GPS MHz Fig. 1: 700MHz band frequency allocation This frequency allocation effectively prevents GPS interference from radiated emissions in cellular applications. However, nearly all modern cellular handsets have internal GPS receivers. Even very weak signals inside the handset can saturate the front-end of these receivers. Despite the guard band, weak intermodulation tails from active adjacent channels may leak into the guard band. With harmonic upconversion, these small signals may cause interference within the handset because of their physical proximity to the GPS receiver. onsequently, the specifications for second harmonic generation in passive components are especially stringent in this band. Previous Investigation Previously [1], we reported on initial experiments aimed at identifying the mechanism that is responsible for these nonlinearities in Silicon Integrated Passive Devices (IPDs). Measured 2nd-harmonc levels in commercial low-pass filters designed for use in the 700MHz band showed little dependence on substrate thickness. Since vertical field strengths are significantly different for different substrate thicknesses, this observed insensitivity led us to suspect that the nonlinear effects giving rise to the harmonics arose mainly through the interac Electronic omponents & Technology onference

3 top electrode bottom electrode increased loss reduced loss fringing field depletion layer (losless) undepleted bulk (lossy) Fig. 2: Proposed nonlinear mechanism Fig. 4: Measured 2nd harmonic power levels in commercial low-pass filters designed for use in the 700MHz band. ence is typical of nonlinear mechanisms that saturate (like, for example, MIS diodes). Fig. 3: Estimated first-order voltage coefficient of capacitance for an example capacitor. tion of lateral electronic fields with the depletion layer in the substrate. This mechanism is illustrated in Figure 2. The results of electromagnetic simulation indicated that for these types of IPD filters, lateral electric fields near the surface are much stronger than the vertical fields. This is especially true near the edges of parallel plate capacitors, as shown in the cross-section in Fig. 2. These lateral fields can modulate the depth of the substrate depletion layers. This proposed mechanism was used to estimate the first-order voltage coefficient of capacitance for a typical example capacitor as a function of substrate resistivity. The key thing to note in this result is that the nonlinearity is expected to fall sharply with increased substrate resistivity. Furthermore, these strong interactions arise at the periphery of capacitors, where the top-level metal passes over the edge of the bottom-level metal capacitor electrode. In many cases, minor design changes can be made to minimize the area of these strong interactions without significantly affecting the IPD s overall electrical characteristics. Substrate Resistivity Dependence Figure 4 shows measured 2nd-harmonic power levels in commercial low-pass filters that are designed to operate in the 700MHz cellular band. These filters had a packaged substrate thickness of 250µm, and were attached to test substrates using bond-wire interconnection. Two samples of each filter were tested to verify the uniformity of the result. t typical cellular power levels, around 20 to 24dm, the filters fabricated on 3.5KΩ-cm substrates showed about 13d lower levels of 2nd-harmonic power. This is consistent with the estimated resistivity dependence suggested by our previous estimates, shown in Fig. 3. t higher power levels, it can be seen that the slope of the curve for the 1KΩ-cm material begins to roll off, and the relative difference between the two substrates is less. This kind of behavior in the power depend- Surface Passivation Experiments It is well-known that high resistivity silicon substrates, without special treatment, exhibit parasitic surface conduction that causes increased conductive losses at high frequency [25]. The losses can be significantly reduced by the formation of a trap-rich surface layer through various methods of ionimplantation or film deposition. More recently, these same surface treatment methods have also been shown to significantly reduce the levels of 2nd-harmonic generation in coplanar transmission line measurements [6, 7]. We have observed an insensitivity of 2nd-harmonic power generation on the device substrate thickness, leading us to propose the model described above for nonlinear conduction driven mainly intense lateral electric fields near the wafer surface. This is also consistent with the other studies linking 2ndharmonic generation to the presence of parasitic surface conduction. The results shown in Fig. 4 suggest that there is also a dependence on substrate resistivity, but since the two types filters compared in that study were fabricated in separate lots, the results are not conclusive. It is possible that there is some difference in the condition of their surfaces, arising from variations in their processing. To help address these questions, we ran further experiments comparing filters with design modifications aimed at reducing the lateral electric field intensity at the substrate surface. Fig. 5 shows maps of the simulated lateral electric field tangential to the substrate surface in these filters. The legend on the right of the figure shows the relative field strength, in d. In the nominal design, strong electric fields (shown in orange) originate at the edges of capacitor plates connected to the filter input and output terminals. In the modified design, the capacitor shape and placement were rearranged to reduce the strength of these fields by nearly 10d. In all other respects, the electrical characteristics of the two filters are nearly indistinguishable. onsequently, we may ascribe any observed difference in the 2nd-harmonic power generation to effects originating near the surface. Fig. 6 shows a comparison of 2nd-harmonic generation in filters fabricated on high resistivity silicon substrates with 1285

4 In addition, in the unpassivated samples we observed that 2nd-harmonic power levels could be increased by several d by illumination, whereas the passivated samples were insensitive. These results suggest that surface conduction mechanisms, either inherent or photo-induced, have nonlinear aspects that contribute to 2nd-harmonic generation. orrelation with other characteristics More recently, we have performed a variety of measurements on IPD wafers of different types. In addition to examining the issues of nonlinear effects, we have also examined loss mechanisms in a number of different test devices. For these experiments, we fabricated IPDs on three different silicon substrate types, listed in Table 1. Substrate types and in these experiments were of very high resistivity silicon, and were available with two different surface treatment methods for passivation. Substrate type was the standard material used in routine IPD production. oplanar transmission lines have been used in several of the past studies cited above to characterize loss and harmonic generation in silicon substrates. The layout of the test device and measured insertion loss for the three wafer types listed above are shown in Fig. 7. These results demonstrate the importance of surface passivation. Note that the insertion loss for lines fabricated on the 1000Ω-cm resistivity wafer () was intermediate between the two 4000Ω-cm substrates having different passivation treatments. The fundamental circuit components in IPD technology are inductors and capacitors. In the useful operating frequency range of IPD circuits, such as filters, diplexers and baluns, the capacitors are relatively high Q devices, and losses are domi- Nominal Design Modified Design Fig. 5: Lateral electric field intensity map for alternative 700MHz-band low-pass filter designs used in this study Table 1: Silicon substrates used for IPD evaluation Substrate Fig. 6: omparison of 2nd-harmonic power generation for unpassivated and passivated substrates, showing the effects of design modification. (shown in blue) and without (shown in red) surface passivation treatment. The results for the nominal design are shown by the solid lines, and those for the modified design are shown by the dotted lines. The substrate wafers in all cases were from a common lot, specified to have resistivity greater than 1000Ωcm, and the filters were made in a common fabrication run. In these results, it can be seen that the design modification resulted in reduction of harmonic power levels of 5 to 6d in the unpassivated samples. This is roughly consistent with the amount of reduction in the lateral field strength, as seen in the plots in Fig. 5. However, in wafers that received surface passivation treatment, there was no appreciable difference in the characteristics of the two designs. Surface conduction is generally known to contribute to harmonic generation. For the unpassivated samples, which presumably have a conductive surface layer, it is not surprising that the modified design, which interacts less strongly with the surface, would show reduced harmonic power. The surprising result is that the passivated samples show similar behavior for both design variants. This suggests that the observed nonlinearities in these devices originate primarily in the bulk Description >4000Ω-cm, surface passivation method 1 >4000Ω-cm, surface passivation method 2 >1000Ω-cm, std surface passivation (control) 3.3 mm Fig. 7: oplanar transmission line test device layout and measured insertion loss.

5 270µm 1.5 mm Fig. 8: Planar spiral inductor test device layout and measured characteristics. nated by the inductors. For this reason, inductor quality factor (Q) is often a figure of merit for technology comparisons. Figure 8 shows the layout of the wafer-probe test inductor used in the evaluation, along with the measured inductance and quality factor for the three substrate types. The trend in these devices, as expected, is the same as in the coplanar transmission lines with higher quality factor correlating with lower transmission line losses. The loss characteristics of IPD circuits show varying degrees of sensitivity to component loss, depending on the type of circuit and design. It can be seen above that substrate loss has negligible effect on the inductance. onsequently, in IPD filters the frequency response is not generally very sensitive to low levels of substrate loss. The main effects are seen in the pass-band insertion loss. For comparison of IPD loss characteristics, we used a common 5th-order low-pass filter design, shown in Fig. 9. This particular filter is designed for operation in the cellular DS band from MHz. The figure shows insertion loss around the center of the pass-band. Here, the same trend as in the above results can be seen. Figure 9 also shows the results of 2nd-harmonic measurements for these devices. In these measurements, the input Fig. 9: 5th-order DS low-pass filter test device layout and measured characteristics. power was at 750MHz and the harmonic was measured at 1500MHz. (This test used the same setup and procedure as the tests for the commercial 700MHz-band filters discussed above.) Note that the harmonic measured results follow the same trend as the loss measurements: Devices fabricated on substrate show the highest insertion losses, and also show the highest 2nd-harmonic power, and devices fabricated on substrate show the lowest losses and lowest harmonic power levels. In an experiment similar to the one described above for the commercial filters, we made design modifications to these DS filters aimed at reducing the surface lateral electric fields. Figure 10 shows a comparison of the original design and the modification. In the original design, the wide horizontal trace 1287

6 The design modifications shown above did not significantly change the filter frequency response. Figure 11 shows the measured 2nd-harmonic power characteristics for the modified design. For comparison, the dashed lines show the results for the standard design. In this comparison, it can be seen that the modified design resulted in reduced levels of 2nd-harmonic power for all three substrate types. In the case of substrate, the measured harmonic levels were very near the noise floor of the experimental setup, so these particular results are less accurate and may be lower than indicated. More remarkably, in the case of substrate the design modification reduced the 2nd-harmonic power levels by more than 20d. s in the earlier experiments using the commercial filters, in these filters we observe photoconductive dependence in the case of substrate. This suggests that the surface passivation method used in this wafer was not completely effective. Typically, methods that create a trap-rich layer on the surface effectively pin the Fermi level and reduce surface mobility. oth of these factors suppress photoconductivity. Table 2 summarizes the results of the various loss and harmonic power measurements. Surface lateral electric field intensity Table 2: Measurements summary measurement PW insertion 5GHz (d) test inductor 5GHz DS filter (d) DS filter 750MHz 2nd-harmonic 24dM input modified DS filter 750MHz 2ndharmonic 24dM input Fig. 8: design modification and simulated surface lateral electric field intensity distributions. Fig. 9: Measured 2nd-harmonic power levels for the modified DS low-pass filter design. across the bottom of the filter forms the ground connection between the input and output pads. This trace is on the top layer of metal, and it also forms the top electrode of several capacitors that are formed beneath this trace, and are consequently exposed to the substrate. In the modified design, the ground trace is made instead using the bottom level metal, and the capacitors lie on top of it. This effectively shields them from interaction with the substrate. s the simulated field intensity distribution maps show, the field intensity is significantly reduced in the modified design. Importantly, these two types of design were located adjacent to each other on the test wafer, so their conditions of bulk resistivity and surface passivation should be nearly identical Discussion The correlation between loss and 2nd-harmonic power generation in silicon IPDs has been noted in past studies, and is generally ascribed to nonlinear effects arising from the semiconductor-insulator interface at the top of the substrate. Very high resistivity substrates tend to behave more like ideal insulators, and consequently these nonlinear mechanisms are less prevalent in them. In our previous study [1] we proposed a mechanism of depletion layer depth modulation by strong lateral electric fields to explain the behavior that we observed, particularly the lack of dependence on substrate thickness. The additional results described above suggest that the mechanism of nonlinear power generation is more complex. s in other studies, we see a direct correlation between loss and 2nd-harmonic power. However, we also observe that high substrate resistivity alone is insufficient to suppress harmonic generation. Surface passivation also plays a very important role. We have seen examples, like the result shown in Figure 6, in which there is no discernible improvement from design modifications. y contrast, we have also seen results like those 1288

7 shown in Figure 11, in which design modifications make a significant difference. These results suggest that both surface losses and bulk losses contribute to 2nd-harmonic power generation in varying degrees depending on the details of design, materials and processing. Side-by-side comparison of design variants provides some useful insight into the relative contributions of surface and bulk nonlinearities. cknowledgements The authors wish to acknowledge the help of Hlaing Ma Phoo Pwint for help with device characterization and Jae Hun Ku and M. Pandi helvam for device fabrication. References 1. R. Frye, K. Liu and R. Melville, Second-Harmonic Nonlinearities in RF Silicon Integrated Passive Devices, Proc. 63rd. IEEE Electronic omponents and Technology onference, May 28-31, 2013, pp H. S. Gamble,. M. rmstrong, S. J. N. Mitchell, Y. Wu, V. F. Fusco, and J... Stewart, Low-loss PW lines on surface stabilized high-resistivity silicon, IEEE Microwave Guided Wave Letter, 9, Oct. 1999, pp E. Valetta, J. van eek,. den Dekke, N. Pulsford, H. F. F. Jos, L.. N. de Vreede, L. K. Nanver, and J. N. urghartz, Design and characterization of integrated passive elements on high-ohmic silicon, in MTT-S Dig., 2, 2003, pp M. Jansman, J. T. M. van eek, M. H. W. M. Van Delden,. L.. M. Kemmeren,. D.. Den. Dekker, and F. P. Widdershoven, Elimination of accumulation charge effects for high-resistivity silicon substrate, in Proc. ESSDER, 3 6, Rong, J. N. urghartz, L. K. Nanver,. Rejaei, and M. van der Zwan, Surface-Passivated High-Resistivity Silicon Substrates for RFIs, IEEE Electron Device Letters, 25, pril 2004, pp D.. Kerr, J. M. Gering, T. G. McKay, M. S. arroll,. Roda Neve, and J. P. Raskin, Identification of RF Harmonic Distortion on Si Substrates and its Reduction Using a Trap-Rich Layer, IEEE Topical Meeting on Silicon Monolithic Integrated ircuits in RF Systems, 2008, pp esar Roda Neve and Jean-Pierre Raskin, RF Harmonic Distortion of PW Lines on HR-Si and Trap-Rich HR-Si Substrates, IEEE Trans. Electron Devices, 59, pril 2012, pp