An Investigation of PCB Radiated Emissions from Simultaneous Switching Noise

Transcription

1 An nvestigation of PCB Radiated Emissions from Simultaneous Switching Noise Sergiu Radu and David Hockanson sun Microsystems, nc. 9 1 San Antonio Road, MS-MPKl 5-12, Palo Alto, CA Abstract: Processors are currently operating with fundamental clock tiequencies that are at or above the resonant frequencies of typical processor boards and modules. Adequately decoupling printed-circuit boards (PCBs) at high frequencies has become an increasingly urgent task in the light of increasing clock frequencies with decreasing rise times. Providing sufficient charge at frequencies near and above 1 GHz is extremely difficult with lumped-element capacitors. To further complicate the issue, modern PCB power buses may be analgous to microstrip-patch antennas. Exciting a power bus at board harmonics may result in significant radiated EM1 from the bus. Much has been done to improve high-frequency decoupling from a signalintegrity perspective. However, the benefit to EM is somewhat unclear, because the mechanism by which power-bus noise results in radiated EM1 is not well understood. nput impedance of a power bus, transfer impedance across a power bus, and radiated emissions from a PCB are presented herein. The results are discussed to provide characterization of radiated EM directly from a PCB power bus.. NTRODUCTON Power-bus noise has become a major concern for EMC engineers. CPU modules may dissipate 8 W or more as a result of processors operating at higher tiequencies and circuits of continually increasing complexity. The DC current draw may be 3 A from a 1.8 V power bus. Consequently, the transient currents are on the order of tens of amps, which results in significant voltage fluctuations on the power bus []. Controlling this problem at the antenna level, and trying to improve the shielding integrity of a chassis beyond 2-3 db at 1 GHz may be expensive, and in some cases mechanically impractical. Modular processor boards are use&l for providing opportunities for customers to upgrade equipment. f a chassis must be significantly improved to meet regulatory emissions limits, the revenue associated with upgrades may be lost. Therefore, significant effort has been given to the problem of designing better power-bus structures [2], [3]. Decoupling is intended to provide locally stored charge during the switching transients of the Cs. Unfortunately, the inductance associated with mounting discrete decoupling capacitors provides a significant barrier to supplying charge at high tiequeucies. To further complicate the matter, there is a great deal of uncertainty as to how much decoupling is sufficient to sufftciently reduce the power-bus noise. The solution lies in knowing the transient currents of the Cs. Data books typically provide average currents, for thermal analysis. Until recently, the chip makers were less interested in analyzing the transient currents, and much more in checking the maximum current densities on the chip, in an effort to prevent electromigration, and to assure a minimum impedance of the chip s power bus, for functionality [4]. For each CMOS circuit, the transient current can be well approximated as a triangular shaped waveform associated with each edge of the clock, or only with one edge of the clock for flip-flop like circuits [S]. Data books provide the powerdissipation capacitance Cm (for thermal calculations) for some CMOS circuits. Cr., combined with information about the output stage (such as R,, and switching time), the current shape, and the capacitive load are adequate for a good estimation of the transient currents [5]. The resulting information derived from the transient current analysis may be sufficient for providing an upper bound to the impedance ofthe power bus. However, the physics behind the radiated emissions from the power bus are not well understood. Reducing power-bus impedance is commonly held to reduce the emissions from a circuit board[q. However, without an understanding of the mechanism by which emissions are coupled or radiated from a power bus, an upper bound for the bus impedance with the goal of low EMl is very difficult to obtain. Radiated emissions from a power bus is gaining acceptance as a possible source of significant EM. A simple PCB was constructed to experimentally investigate the possibility of radiated emissions directly from a power bus. nput impedance and ]Szl( were measured on the PCBs and contrasted with radiated emissions measurements. To further support the conclusions from the experimental boards, similar measurements were conducted on bare fabrication and populated production-level processor boards.. RADATON MECHANSMS There are a number of theories as to how radiated emissions result from PCBs. Many theories are supported through experience, and virtually all are difficult to argue against. Consequently, it is difficult to determine in a general fashion what the dominant EM mechanisms in a PCB are. Possible radiated EMl could result from: a) direct radiation from the traces (top/bottom microstrip structures), b) dipole-like radiation from a PCB with attached cables, where the source driving the dipole is the result of a current driven mechanism, c) parallel- or perpendicular-module conligurations, where an equivalent noise source is generated at a module connector and the daughter card modules are driven against a motherboard, d) direct radiation Tom Cs and heatsinks, e) or direct radiation from the power bus. ln the shadow of items (a) through (d), direct radiation from the power bus appears to be a second or third order phenomenon, as opposed to a primary source of radiated EM. n particular when the power bus has plane separations of 2-4 mils, given that a microstrip patch antenna must typically have planar separations of approximately 4-6 mils to be effective. However, there are some cases where, by 893

2 2 8 1 ii ;, 1 ~ ; f-----f ;:! / #! ~g:&g Figure 1. nput impedance results for a power bus with a 2 mil separtion. the process ofelimination, EMC engineers are left with the conclusion that radiating power-buses are the only option, such as might be the case with hand-held computing devices.. EXPERMENTAL RESULTS The relevance of radiating power-bus structures was investigated experimentally. Two PCBs were constructed from double-sided, 9 x12 copper-clad boards. The boards were Rogers RO435 microwave laminate boards with a releative dielectric constant ~,=3.48. The two boards were identical except that one was 2 mils thick and the other was 6 mils thick. Three 3Smm SMA connectors were mountd on each board. One was mounted in the center of the planes, one in a corner (1 away Tom each edge), and one in the middle of the nine inch edge (1 from the edge). The connectors were mounted on the ground plane, and the center conductors penetrated the board to connect to the power plane. An HP8753C Network Analyzer was used to measure Sl 1 at each port, and converted to input impedance. Furthermore, the network analyzer was used to measure S21 between two ports on the experimental boards. Radiated emissions were measured in an agency-certified semi-anechoic 1 meter chamber using Rohde & Schwarz EM1 Receiver. The board was excited with the tracking generator of the EM1 Receiver, and emissions were measured with a biconilog antenna Above 1 GHz a horn antenna was employed. The correction factors were included in the data, with the exception ofthe cable losses between the EUT and the tracking generator. The cable losses were approximately 4 db at 1 GHz and 9 del at 5 GHz. Consequently, the radiated emissions values are only relative values and A-B comparisons are made between results for different configurations, as opposed to real measurements that might be compared to regulatory limits. The power bus structure is similar to a microstrip-patch antenna The resonance frequencies can then be calculated using a cavity model. The thickness of the boards is very small, and for the purposes of calculating resonance tiequencies the field distribution can be approximated as constant between the planes along the axis -1-2 ;; 8-3 m -4 Figure 2. nput impedance results for a power bus with a 6 mil separtion.._ -i - 2 mik, comer-cd --,gle l j. j Figure 3. S211 results for a power bus with a 2 mil separation. normal to the planes. The boards have the same L, W dimensions, and therefore must have the same resonance frequencies. The expected resonant frequencies under 2GHz are: 412.9MHz, 548.8MHz, 686.8MHz, 825.7MHz, 991.5MHz, 198MHz, 1173MHz, 1239MHz, 1355MHz, 1374MHz, 1646MHz, 1655MHz, 1697MHz, and 1842MHz The input impedance for a 2 mil separation and 6 mil separation are shown in Figures 1 and 2, respectively. Results are shown for an input port centered on the board, and an input port near the corner of the board. The resonant frequencies are approximately the same for both boards, supporting the hypothesis that the resonant frequency is independent of the board thickness. The resonances calculated at M& MHz, and MHz are evident in the measurement results for the geometry where the input port is 894

3 Figure 6. Radiated emissions results for a power bus with a 2 mil separation, a centered source, and the 9 side facing the antenna. Figure 4. S211 results between a port located near the corner and a port located near the edge for a 2 mil and 6 mil power-bus separation m 8 1 Eresuency W-w Figure 5. S211 results for a power bus with a 6 mil separation. centered on the PCB, and the measured and calculated frequencies show good agreement (-3% deviation). nterestingly, the resonance at MHz is not apparent in Figures 1 or 2 for the geometry with a centered source. A reason for the lack of excitation has yet to be determined, as the source location is optimum for this mode. Moving the source location to the corner results in the addition of six resonance frequencies. The resonance at MHz was predicted with the cavity model, however the remaining resonances were not. The resonance tiequencies not predicted by the cavity model indicate that a transmission-line model is more appropriate for determining the resonance tiequencies. Locating the source port near the corner results in excitation of the planes 11 and 8 t?om the edge ofthe board. Consequently, quarter-wave transmission-line resonances associated with 11 and 8 correspond to resonances at 226 MHz, and Figure 7. Radiated emissions results for a power bus with a 2 mil separation, a centered source, and the 12 side facing the antenna Figure 8. Radiated emissions results for a power bus with a 2 mil separation, a source located near the corner, and the 12 side facing the antenna. 895

4 _- '.., j 291OOiOO Figure 9. High-frequency radiated emissions results for a Figure 11. Radiated emissions results for a power bus power bus with a 2 mil separation, a centered source, with a 6 mil separation, a centered source, and the 12 side facing the antenna. and the 12 side facing the antenna s o Figure 1. Radiated emissions results for a power bus Figure 12. JSigh-frequency radiated emissions results with a 6 mil separation, a centered source, for a power bus with a 6 mil separation, a centered and the 9 side facing the antenna. source, and the 12 side facing the antenna. 311 MHZ, respectively. These two resonances agree well with the resonances measured for the experimental board with the source port located near the corner, as shown in Figures 1 and 2. S211 was also measured on the experimental boards. The results for a 2 mil planar separation is shown in Figure 3 for measurements between ports located at the corner and the center, and ports located between the comer and the edge. The results indicate a dependence on port location. Although all the resonances shown in Figure 1 were measured in S21 1, the level of the resonances was very dependent on port location. Furthermore, the resonance at MHz is evident in Figure 3, although it was not evident in Figure 1. The results for S211 between the comer and the edge are contrasted fw2 mil and 6 mil separation in Figure 4. The interplane capacitance is three times greater for the 2 mil separation than the 6 mil separation. This correlates very well with the 1 db decrease in S211 measurements shown in Figure 4. The micro&@-patch antenna analogy assumes that the emissions may be predicted by modeling the edges of the PCB with magnetic-current density vectors. Consequently, modifying the edges ofthe board should have an impact on the input impedance. Wrapping the edge of the PCB with copper tape (while maintaining DC isolation between the ground plane and the power plane) should be aualgous to capacitively loading the parallel-plate transmission-line. The experimental board with a 6 mil interplane separation was modified by attaching copper tape to the ground plane. The tape was folded over the edges and extended 2 inside the edge over the power plane. The paper backing on the copper tape was used to provide DC isolation between the tape and the power plane. S211 results for the modified and unmodified models are shown in Figure 5. The resonance frequencies were shifted by approximately 2%- Radiated emissions were also measured in a semi-anechoic 1 meter chamber. The emissions for horizontal polarization were uniformly 2 db below the results for vertical polarization, therefore only the results for vertical polarization (antenna parallel to the normal vector of the planar structure) are given herein. The units of the emissions results shown herein are dbuv, and the kequency is given in h4hz. The excitation was varied, and the emissions from the 12 and 9 long sides were measured The results for the 2 mil board for center excitation and the 9 side facing the antenna, center excitation 896

5 and the 12 side facing the antenna, comer excitation and the 12 side facing the antema, and high-frequency (l-5 GHz) emissions with center excitation and the 12 side facing the antenna are shown in Figures 6, 7, 8, and 9, respectively. The peaks in Figures 6 are at the same frequencies as the peaks in the input impedance1 for the same board and the same source location (see Figure 1). The resonance frequency at MHz is associated with a field distribution over the 9 dimension. The field distribution along the 12 edge is assumed approximately constant as a result of the cavity model. Replacing the edge fields with magnetic-current densities results in the sum of emissions Tom two current densities that are located spatially 18 degrees apart. Consequently, Figure 6 shows a high emission near 55 MHz, while Figure 7 shows a significantly lower emission near 55 MHz. n contrast, the resonance at 686 MHz is associated with a field distribution along the 12 side of the PCB, and a high emission is measured near 69 MHz, as shown in Figure 7. The similitude of the radiation pattern with the input-impedance pattern is further supported by the results shown in Figure 8, when compared with the results shown in Figure 1, where the board is excited in the comer. Radiated emissions results for the fkequency range between l-5 GHz are shown in Figure 9. These results are provided to demonstrate the possibility of emissions related to board resonances through increasingly higher tiequencies. Radiated emissions were also measured for the 6 mil board for a centered source with the 9 side facing the antenna, a centered source with the 12 side facing the antenna, and emissions results for the frequency range between l-5 GHz with a centered source and the 12 side facing the antenna, and the results are given in Figures 1, 11, and 12, respectively. The trends evident in Figures 1, 11, and 12, correlate with those of Figures 6, 7, and 9, respectively. The significant difference between these two sets of results is that the emissions at the peaks for the 6 mil board are 1 db higher than the those of the 2 mil board. The 1 db reduction corresponds to the factor of three difference in the separation between the planes. Further experiments were performed involving shielding the edges with copper tape. The greater the overlap of the ground plane over the power plane, the more the peaks were diminished. A 3 db attenuation was achieved by shielding all but a 2 W area near the center of the power plane, however the peaks were still evident above the noise floor. The results of these experiments indicate that the quality factor or Q of these resonances is high. Hypothetically, loss may be employed on the power bus, however, to reduce the Q, while still maintaining functionality, as will be discussed in Section V. J.V. NUMERCAL RESULTS The input impedance of a power bus geometry was also investigated numerically. A power bus structure oftwo PEC planes of 1 cm x16 cm, with a separation h=2 mm was modeled using FDTD. The dielectric constant was that of air (&,=). The 5 Ohms, 1V excitation source was located alternately, in the geometrical center of the planes, and near a comer (4 cm?om a 16 cm side, and 2 cm from a 1 cm side). Cubic cells were employed to discretize the region, and the time step was set to dt=dl/2c, where dl = 1 mm was the length of one side of a cubic cell, and c is the speed of light in free space. Eight layer PML absorbing boundaries were used and located a minimum ofeight cells away from the model. Amodulated sincsource was modeled, and the data was Fourier transformed to provide input impedance results from OOMHz to 5GHz. The results are presented 1. d 1. 8 ;/ 1 $ i 7 1; $ s,---&-j -,------r T; L - ---A! * it, ;: i. :: : 1; <> ~: ; ;! : -:,,-, ii, i ii A -;.--7r-!-!;. $ :i c, 4 ji i 1 : * $ :, - ;- i 1 r :, t 4-4 j,i, i _ , -ZL - / ;fi _-.. f T , j : : --,----~ , L------L---~- yi j j ; ; +a&-- CLO...b,,..q., a,,,.1 1 2oOo 3ooo 5ooo Fmquency (M&z) Figure 13. nput impedance results computed for a 16 cm x 1 cm power bus with a 2 mm separation. in Figure 13. For a cavity with the same dimensions as the model the first resonances are expected at 936.9MHG 1499MHz, 1768MHz, 1874MHz, 24OOMHz, etc. These resonances were found with less than 3% deviation from the FDTD simulation. The results shown in Figure 13 also support the transmission-line model of the power bus. Locating the source near the comer results in more resonance frequencies, because of the increase in the distance i?om the source to the edge of the board. V. CASE STUDY: A CPU MODULE UPGRADE The use of CPU modules in computers provide for economical upgradability (processor type, speed, L2 cache memory), as well as making more efficient use of available real-estate. A new module designed to go in a shipping product resulted in high emissions near 9 MHz, which was the second harmonic of the processor. The new module was significantly different from the original modules, as a result of a die-shrink, and a new layout. Problems at this li-equency were not completely unexpected, because the internal PLL clocked to 9 MHz and a-divide-by-two scheme was employed to obtain a 45OMHz clock with a 5% duty cycle for the core logic. Also, the transient currents associated with the core logic are generated on both edges of the clock, resulting in power bus noise at double the fundamental clock frequency. However, the 9OOMHz was unusual strong, indicating a resonance at the chassis level or at the CPU module level. Preliminary investigations suggested that the 9 MHZ radiated from the module by two mechanisms: direct edge radiation, but also the vertical mounted CPU module contributed to radiated EM1 as a monopole antenna driven against the motherboard by a power bus related equivalent source at the connector level. The original module was constructed with 14 layers. The noise sources were determined to be power bus related, and a 16 layer board was constructed to determine if the emissions could be reduce with a lower-impedance power-bus. S211 was measured on 14 layer and 16 layer modules, populated and unpopulated, with the network analyzer. The results for the four configurations are shown in Figure 14. The results for the bare board show a strong resonance at approximately 9 MHz. Adding extra layers, and reducing the f, 897

6 V. CONCLUSON g -7 g Layers, bareboard Layers, bareboard t -11 _- _ / n Frequency N-W Figure 14. S211 results for a production-level PCB power bus. separation between the power planes, resulted in approximately twice the interplane capacitance. Consequently, the bare board S211 measurements for the 16 layer board were approximately 1 db lower than the 14 layer board, as shown in Figure 14. The radiated emissions for the final system were 6 db lower with the 16 layer board, than with the 14 layer board. The PCB was also used for a4 MHzmodule. As with the 45 MHzmodule, significant emissions were measured at the second harmonic. However, there is not a peak in the (S21( measurements shown in Figure 14. Changing to a 16 layer board resulted in a 6 dl3 reduction in the emissions from the 4 MHz module as well. n Section l, the emissions results were found to correlate to the input impedance results, and not necessarily to the JS211 results. Consequently, S211 results may not be sufficient in predicting problem frequencies. A board with loss on the power bus may not support field levels as high as a simple copper clad board, or unloaded PCB. nput impedance measurements were not performed on the production-level modules, because of the difficult associated with calibrating down to the board without a PCB mounted connector. The S211 results for the bare board configurations were relatively high Q. A populated board includes losses on the power bus as a result of the very dense population of discrete components. The S211 results for a populated board (14 and 16 layers) are shown in Figure 14. The level of the S211 measurements were significantly lower than the bare board results. Beneficial decoupling strategies have been presented in EMC literature for providing low-impedance power-buses for improved signal integrity. The effect on radiated EM has also been investigate, although the mechanism by which power bus noise results in radiated EM is not, in general, well understood. Power-bus structures are analgous to microstrip-patch antennas, and the possibility that powerbuses may radiated as such has received some attention. Power-bus radiation is investigated herein using experimental and numerical models. A case study is also discussed to provide support for the experimental results. Using simple copper-clad boards, high-q resonances in input impedance impedance and S211 results were measured. A simple cavity model was used to predict resonance frequencies, and found to be inadequate for determining all the resonance frequencies associated with exciting the power bus at a given point. A transmission-line model proved more accurate over the frequency range of interest. FDTD simulations were used to validate the transmission-line approach. Radiated emissions were also measured Tom the experimental boards. The input impedance resonances correlated well with the radiated emissions, while the S21 measurements were found to miss peaks. A case study was reviewed to show the reduction in emissions in a production-level device, as a result of lowering the power-bus impedance. REFERENCES [] Senthinathan, R., Price, J., Simultaneous Switching Noise of CMOS Devices and Systems, Kluwer Academic Publishers, Boston, [2] Kashyap, N., Hubmg, T., Drewniak, J., Van Doren, T., An Expert System for Predicting Radiated EM1 from PCBs, in Proc. EEE nt. Symp. On EMC, pp , Austin, Texas, August 18-22, [3] Costa, V., Preatoni, R, Caniggia, S., nvestigation of EM on Multilayer Printed Circuit Boards: C - noise and Power Supply Decoupling, in Proc. EEE nt. Symp. On EMC, pp , Santa Clara, CA, August 19-23, 1996 [4] Coenen, M., On-Chip Measures to Achieve EMC, in Proc. of 12* nternational Zurich Symposium on Electromagnetic Compatibility, pp.3 l-36, Zurich, Switzerland, February 18-2, [5] S. Radu, R. E. DuBroff,, T. H. Hubing, T. P. Van Doren, Designing Power Bus Decoupling for CMOS Devices, in Proc. EEE nternational Symposium on FMC, pp , Denver, Colorado, U. S. A., August 24-28, ]6] Rainal, A J., Computing nductive Noise of CMOS Drivers, in EEE Transactions on Components, Packaging, and Manufacturing Technology - Part B, pp , vol. 19, No. 4, November