146

IEEE TRANSACTIONS ON ELECTROMAGNETIC COMPATIBILITY, VOL. 41, NO. 2, MAY 1999

[5] D. Levin, Development of nonlinear transformations for improving convergence of sequences, Int. J. Comput. Math. (B), vol. 3, pp. 371388, 1973.

Measuring the Radiated Emissions from a Family of Microprocessors Using a 1-GHz TEM Cell
Kevin P. Slattery, James P. Muccioli, and Terry North

Fig. 4. Looploop shielding effectiveness, radius 1 KHz.

= 0 05 m, 1/16 copper,
:

AbstractThis paper presents a series of measurements of the radiated emissions from 8 and 16 bit microprocessors. The radiated emissions were measured using a 1-GHz TEM cell that incorporates the device under test (DUT) into the cell structure itself [1]. For the 16-bit processor, samples from each of the manufacturers identied process corners were measured and compared. Two separate fabrication lines were compared for process variability. Spatial location on the wafer was measured for emissions variation. In addition, emissions were measured for a 16-bit processor as a function of operating temperature. Finally, a comparison was made between discrete implementations of a module digital core and the same circuit implemented as an multichip module (MCM). Index TermsMicroprocessor, radiated emissions.

I. INTRODUCTION The 1-GHz Fischer TEM cell was developed expressly for the purpose of measuring the emissions from very large scale integration (VLSI) devices in the absence of printed circuit board (PCB) effects. The TEM cell used is described in Society of Automotive Engineers (SAE) specication 1752/3, Integrated Circuit Radiated Emissions Measurement Procedure and is manufactured by Fischer Custom Communications. Essentially, the Fischer TEM cell incorporates the device to be measured, a microprocessor, or any large scale device, into the TEM cell structure itself. By doing so, the device under test has negligible effect upon the cell characteristics [1], [2]. The emissions measured by the TEM cell have been shown to correlate with emissions as measured in a semi-anechoic chamber [1]. The TEM cell is single moded to 1.3 GHz. The devices under test (DUTs) measured and described in this paper were manufactured by Motorola and are typical of devices found in automotive control electronics. Each processor is mounted on a support board that is then placed in the TEM cell (Fig. 1). All of the DUT support circuitry resides outside the cell; only the DUT is inside. The circuit board itself becomes a part of the cell oor. The PCB is of multilayer construction. The outer layer under the DUT is ground. II. MEASUREMENT PROCEDURE AND REPEATABILITY The repeatability of the Fischer TEM cell was tested at ten different bands in the frequency range. At each point the device was swept 20 consecutive times and the maximum peak was noted. Fig. 2 shows the maxmin delta for these points. Repeatability was seen to be
Manuscript received May 1, 1997; revised February 5, 1999. K. P. Slattery is with Recardo North America Inc., Madison, AL 35758 USA. J. P. Muccioli and T. North are with the Chrysler Corporation, Auburn Hills, MI 48326 USA. Publisher Item Identier S 0018-9375(99)04330-6.

Fig. 5. Looploop shielding effectiveness, radius = 0:5 m, 1/16 copper, 1 KHz.

integration over the receiving loop must be included. This provides a monotonic behavior of SE with looploop spacing. With modern computers and integrators, numerical results are readily obtained. ACKNOWLEDGMENT The authors would like to thank Dr. T. Griffy for his suggestions. REFERENCES
[1] J. R. Moser, Low-frequency shielding of a circular loop electromagnetic eld source, IEEE Trans. Electromagn. Compat., vol. EMC-9, pp. 618, Mar. 1967. [2] C. M. Ryan, Computer expression for predicting shielding effectiveness for the low-frequency plane shield case, IEEE Trans. Electromagn. Compat., vol. EMC-9, pp. 8394, Sept. 1967. [3] P. R. Bannister, New theoretical expressions for predicting shielding effectiveness for the plane shield case, IEEE Trans. Electromagn. Compat., vol. EMC-10, pp. 27, Mar. 1968. [4] , Further notes for predicting shielding effectiveness for the plane shield case, IEEE Trans. Electromagn. Compat., vol. EMC-11, pp. 5053, May 1969.

00189375/99$10.00 1999 IEEE

Authorized licensed use limited to: ULAKBIM UASL - GEBZE INST OF TECH LIBRARY. Downloaded on May 2, 2009 at 11:03 from IEEE Xplore. Restrictions apply.

IEEE TRANSACTIONS ON ELECTROMAGNETIC COMPATIBILITY, VOL. 41, NO. 2, MAY 1999

147

Fig. 3. Spectrum from discrete digital core.

Fig. 1. Side view of Fischer TEM cell.

Fig. 4. Spectrum from MCM version of digital core.

Fig. 2. Vertical axis is decibels; horizontal axis is in megahertz.

quite good and generally less than 2 dB of total variation was seen. Standard deviation was seen to be on the order of 0.6 dB. The noise oor for the measurements is given as the noise oor for the instrumentation usedin this case, as 0121 dBm. All measurements were performed over the span 11000 MHz using a 32-dB Sonoma preamp and an HP 8568B spectrum analyzer. Sweep time was set at 150 s. This was chosen so that the digital core would execute at least one cycle of its program code at each spectral sample point (1001 points for the HP 8568B). Each processor was programmed to run functional production code. The digital core in this instance is running at a fundamental frequency of 16 MHz. The following plot shows the preceding spectrums after they have been smoothed with a 16point moving average. This procedure allows for a more direct comparison of devices by removing spectral clutter. It should be noted that strong harmonics below approximately 150 MHz are also removed or reduced. In comparing the multichip module (MCM) and discrete spectrums (Figs. 3 and 4), one can see that the magnitude of the strongest spectral peak below 150 MHz is approximately the same for both types. The smoothed spectrums, by removing extreme spectral variation, show quite dramatically the differences between the two implementations of the same function (Fig. 5). Fig. 6 shows a comparison of the full spectral data with a smoothed version of the same data set.

Fig. 5. Comparison of ltered spectrums for MCM and discrete versions.

III. RADIATED EMISSIONS FROM MICROPROCESSORS A. Emissions from 8-Bit Processors The following graphs show a comparison of 8-bit processors running two different sets of program code. The code measured in these graphs represents typical production code for body controller electronics.

Authorized licensed use limited to: ULAKBIM UASL - GEBZE INST OF TECH LIBRARY. Downloaded on May 2, 2009 at 11:03 from IEEE Xplore. Restrictions apply.

148

IEEE TRANSACTIONS ON ELECTROMAGNETIC COMPATIBILITY, VOL. 41, NO. 2, MAY 1999

Fig. 6. Comparing the unltered spectrum with the ltered spectrum. Fig. 8. Spectral differences due to software variations.

Fig. 7. Comparison of smoothed spectrums from 8-bit processors.

In Fig. 7, we see a comparison of the HC11 and HC12 processors. The HC12 is a later version of the HC11. Fig. 8 shows that the TEM cell measurement procedure can indicate variation in emissions levels due solely to software variation. The dark line shows a measure of the emissions for an earlier version of software. The lighter dashed line shows a more recent software level. One observes that the emissions have increased from 50 to 550 MHz. The third line shows the emissions due to software from a different product that uses the same processor. All of these emissions measurements were taken using the exact same processor the HC12 but programmed with the three different sets of code. Fig. 9 shows a comparison of the smoothed spectra for two 8-bit processors and for a 16-bit processor. As is apparent from the plot, the jump in emissions levels is quite dramatic as one goes from 8 to 16 bit. The 16-bit processor has signicant levels (>10 dBuV) to 700 MHz.

Fig. 9. Comparing the spectrums produced by 8- and 16-bit processors.

IV. COMPARISON OF PROCESS VARIATIONS IN A 16-BIT PROCESSOR During the course of normal module-level emissions testing, it was found that repeatability was not as good as it should have been and indeed had been for previous modules of the same type. After a lengthy process of investigation it was discovered that the cause of the discrepancies appeared to be associated with differences between processors fabricated at one assembly plant versus another plant. In order to ascertain whether this was indeed the reason, the manufacturer supplied us with several processors identied with each

Fig. 10. Comparison of four 16-bit processors from fab site 1.

point of process variation from each of the plants that manufactured the particular processor. The following graphs show the results. The four measured variations in process were: 1) p-channel thresholdLow/High; 2) n-channel thresholdLow/High; 3) effective transistor widthLow/High; 4) effective transistor lengthLow/High.

Authorized licensed use limited to: ULAKBIM UASL - GEBZE INST OF TECH LIBRARY. Downloaded on May 2, 2009 at 11:03 from IEEE Xplore. Restrictions apply.

IEEE TRANSACTIONS ON ELECTROMAGNETIC COMPATIBILITY, VOL. 41, NO. 2, MAY 1999

149

Fig. 11.

Comparison of four 16-bit processors from fab site 2. Fig. 12. Comparison of three proccessors from the same process corner.

From these parameters, the following corners were dened: 1) corner 1: LLLL; 2) corner 2: HLHH; 3) corner 3: LHHH; 4) corner 4: HHHH. The averages across the measured spectrum for each of the process corners are (3G26): 1) process corner 1: 16.9 dBV; 2) process corner 2: 16.3 dBV; 3) process corner 3: 17.5 dBV; 4) process corner 4: 20.0 dBV. It is seen that process corner 4 stands out from the others in terms of a higher average, but process corner 2 is different as well, since its spectrum is apparently shifted down in frequency compared to the other three corners. Process corner 2 has a higher n-channel threshold so one would expect that the switching speed would be slower than the others and, hence, would have a lower frequency distribution than the others. Process corner 4 has a higher n-channel threshold as well, but in addition has a higher p-channel threshold. Perhaps symmetry is involved in explaining why one of the process corners is lower in frequency versus the other. However, Fig. 13 is a comparison of 3 three dies from the same process corner and shows only one of the dies being shifted lower in frequency. Fig. 11 shows a comparison of the process corners for the processor manufactured at fab site 2. The averages across the measured spectrum for each of the process corners are (1G73): 1) process corner 1: 18.3 dBV; 2) process corner 2: 19.8 dBV; 3) process corner 3: 18.2 dBV; 4) process corner 4: 17.5 dBV. For fab site 2, no single process corner stands out as dramatically as for fab site 1 (see Fig. 10). The next graph shows the variation for a given processor within the same process corner. For the rst process corner shown in Fig. 12, variation across dies is restricted to f > 650 MHz. For the corner shown in Fig. 13, it is associated with only one die and is below 200 MHz. We also measured the variation in measured emissions as a function of the spatial location of a die in the process on the same fab line (3G26). The spectral variation due solely to the processors location on the wafer is rather small (Fig. 14). When we look at the variation as a function of process corner and fab line, we see that fab site 1 has a larger variation than does fab site 2: Average delta, MaxMin:

Fig. 13. Comparison of process variation within a selected die group.

Fig. 14. Spectral variation due to die location on wafer.

1) fab site 1: 5.1 dB; 2) fab site 2: 3.9 dB; 3) fab site 1 spatial variation: 1.7 dB. V. SOME INFERENCES
FROM THE

MEASURED EMISSIONS DATA

The following plot shows a smoothed graph from one of the 16bit processors from the 3G26 lot (Fig. 15). Comparison of all the emissions plots from all the processors show similarities in that there appears to be two distinct break points in the data sets. The rst break point occurs at approximately 156 MHz and is associated

Authorized licensed use limited to: ULAKBIM UASL - GEBZE INST OF TECH LIBRARY. Downloaded on May 2, 2009 at 11:03 from IEEE Xplore. Restrictions apply.

150

IEEE TRANSACTIONS ON ELECTROMAGNETIC COMPATIBILITY, VOL. 41, NO. 2, MAY 1999

Fig. 16. Comparing the spectrums for MCM versus discrete digital core. Fig. 15. Smoothed spectrum for 16-bit microprocessor.

with the measured input/output (I/O) rise time, 45 ns. The second break point occurs at approximately 550600 MHz and is apparently associated with the internal die rise time, given by the manufacturer as 500700 ps. The data also shows that the spectral region from 150 to 600 MHz remains fairly at, showing no signicant falloff in emission levels. Additionally, we see that the emissions below 150 MHz are dominated by clock associated harmonics. The emissions above 150 MHz are somewhat more problematic and require further study. As a tool for measuring process variation dependence in relation to radiated emissions, the TEM cell method appears to be unique. It is repeatable and reliable and can be shown to be related to actual emissions measured in modules using the DUT [1]. VI. COMPARISON OF DISCRETE DIGITAL CORES WITH MCM IMPLEMENTATION The following graph shows a comparison of the digital core electronics of an engine controller implemented on the TEM cell test board in both a discrete version and an MCM version. The discrete version consists of the processor, application-specic integrated circuit (ASIC) and ash memory. The MCM uses the exact same parts and is fabricated on the same substrate. Refer to Figs. 3 and 4, for the complete spectrums. Average improvement between the discrete implementation and the MCM is 11 dB. As Fig. 18 shows, the MCM has a low measure of variability. In Fig. 16, it is interesting to note the extremely at emissions of the MCM from 50 to 550 MHz. VII. ORIENTATION VARIATION BETWEEN MCM
AND

Fig. 17. The difference in decibels between the MCM and the discrete implemetation.

DISCRETE

The following graphs show the measured differences for the discrete and the MCM when the TEM cell test board is rotated through 90 in the TEM cell. The MCM version shows a much lower degree of orientation dependence than does the discrete version. This is partially explained by realizing that the MCM version occupies a signicantly smaller PCB area than does the discrete and, therefore, has oriented structures that are smaller. However, this is only a partial explanation, since the major radiating structure is associated with the processor itself; that is seen to be the internal bus structure of the processor and that structures size does not change from discrete to MCM.

Fig. 18. Comparing spectral variations between three MCMs.

As Fig. 20 shows, the discrete version starts showing signicant variation above 500 MHz while the MCM shows smaller variation with orientation at higher frequencies. As noted previously, the strong clock harmonics below 200 MHz are attenuated when the moving average is taken and, hence, the strong spikes are absent

Authorized licensed use limited to: ULAKBIM UASL - GEBZE INST OF TECH LIBRARY. Downloaded on May 2, 2009 at 11:03 from IEEE Xplore. Restrictions apply.

IEEE TRANSACTIONS ON ELECTROMAGNETIC COMPATIBILITY, VOL. 41, NO. 2, MAY 1999

151

Fig. 21. Comparison of orientation variation between MCM and discrete. Fig. 19. Spectral response as a function of orientation in the TEM cell.

Fig. 22. The spectrum of the HC16 at 25  C.

Fig. 20.

Rotational variation of discrete versus MCM.

(it should be noted that the moving average removes strong narrow peaks, and was only used as a means to compare complex spectra). Under rotation, these harmonic spikes are seen to vary by 1015 dB. This strongly suggests that they are the result of directed currents in the processor structure, associated with the internal processor bus. Additionally, it was noted that two rotations were sufcient to characterize the rotational dependence of the emissions [3]. Fig. 21 shows a comparison of the rotational differences for the two types of implementation and overall indicates very little real difference in terms of the maximum observed differences. We take this to show that the processor, identical in both cases, is the source of the major emissions and that those emissions are due almost entirely to the processor and not to the circuit connections to the ASIC and the ash memory.

Fig. 23. Spectrum of the HC16 at 100  C.

It is interesting to note that as the temperature increases, the emissions spectra moves up in frequency and down in amplitude and as the temperature decreases, the spectra moves down in frequency and up in amplitude. IX. SUMMARY The Fischer TEM cell has shown itself to be a versatile and remarkable device for the characterization of processor emissions. The repeatability of the device is good, the speed of measurements is fast. This has allowed the authors to perform a large number of measurements of microprocessors that previously were not possible. The end result of these measurements has been to give us a better look

VIII. MEASURED EMISSIONS AS A FUNCTION OF TEMPERATURE A convenience of the Fischer cell is its size. This allowed us to place the entire cell in one of our environmental chambers and then to vary the operating temperature and measure the resulting emissions. Figs. 2224 show the ambient spectrum, the hot spectrum, and the cold spectrum. Fig. 25 shows the results when the preceding three spectrums are smoothed with a 16-point moving average and compared.

Authorized licensed use limited to: ULAKBIM UASL - GEBZE INST OF TECH LIBRARY. Downloaded on May 2, 2009 at 11:03 from IEEE Xplore. Restrictions apply.

152

IEEE TRANSACTIONS ON ELECTROMAGNETIC COMPATIBILITY, VOL. 41, NO. 2, MAY 1999

ACKNOWLEDGMENT The authors would like to thank the people at Motorola for their uninching support given during the gathering of these measurements. They would also like to thank D. Graden and S. Merrill. REFERENCES
[1] J. Muccioli, T. North, and K. Slattery, Investigation of the theoretical basis for using a 1 GHz TEM cell to evaluate the radiated emissions from integrated circuits, presented at IEEE Int. Symp. Electromagn. Compat., Santa Clara, CA, Aug. 1996. [2] A. Engel, Model of IC emissions into a TEM cell, presented at IEEE Int. Symp. Electromagn. Compat., Austin, TX, Aug. 1997. [3] T. DiBene and J. Knighten, Effects of device variations on the EMI potential of high speed digital integrated circuits, presented at IEEE Int. Symp. Electromagn. Compat., Austin, TX, Aug. 1997.

Fig. 24.

Spectrum of the HC16 at

040  C.

A Simple Formula for the Ground Resistance Calculation

Sergey L. Loyka

AbstractSimple approximate formulas for the ground resistance and the potential distribution that have quite good accuracy are presented. The use of these formulas can simplify and make faster the calculation of ground path coupling. Index TermsGround-path coupling, ground resistance, potential distribution.

I. INTRODUCTION A common ground path is a way of coupling between a disturbing source and a victim circuit. This kind of coupling has got a detailed consideration in the literature. The resistance between electrodes and the potential distribution are important quantities for the coupling calculation [1]. The resistance between the two electrodes on a nitely conducting plate of thickness t and of innite extent is given by [1]

Fig. 25.

Spectral emissions as a function of temperature.

into the structure of emissions from VLSI devices and has allowed us to begin building a database. The creation of this database will allow us to be able to compare any and all changes that are made at the device level and at the software level. Additionally, observation of so many spectrums has enabled us to begin speculating about possible analytical models for describing the emissions from VLSI devices. The measured spectrums are seen to have rapid variations across the measured band. Two explanations come to mind. The rst is that the spectrum is a result of an ensemble of pseudorandom binary processes. The second is that the spectrum is the result of an excited two-dimensional electron gas in the underlying substrates of the device. The second explanation is described in the companion paper to this paper. It has the merit that it explains the temperature dependencies of the measured spectrums while the rst explanation cannot. The measured levels of the 16-bit devices are signicant for electromagnetic interference. The levels are high enough for coupling to occur between the processor and other regions of the PCB. This has been seen to occur in certain automotive applications. The processor has been seen to be the source of common mode currents that have been coupled to the module I/O connector. These currents in turn, drive the electrical system harness.

Rab =

 a

1
+2

n=1

1 2 + (2nt)2 a

1 2 + (2nt)2 d

(1) where a is the electrode radius, d is the separation between the two electrodes, and  is the plate conductivity. The geometry of the problem is shown on Fig. 1. As it is pointed out in [1], .. . .each term in the sum behaves asymptotically as 1=n and, consequently, is not strictly summable independently. However, the two terms in the sum, when taken together, fall off faster than 1=n, and the innite series converges. To determine the exact degree of the falling off, we bring the two terms in the sum to the common denominator and multiply the numerator and the denominator by the same expression
Manuscript received October 14, 1998; revised February 5, 1999. The author is with the Swiss Federal Institute of Technology, LEMA-EPFLEcublens, Lausanne, CH-1015 Switzerland. Publisher Item Identier S 0018-9375(99)04088-0.

00189375/99$10.00 1999 IEEE

Authorized licensed use limited to: ULAKBIM UASL - GEBZE INST OF TECH LIBRARY. Downloaded on May 2, 2009 at 11:03 from IEEE Xplore. Restrictions apply.