From IC characterization to system simulation by systematic modeling bottom up approach Frédéric Lafon, François de Daran VALEO VIC, Rue Fernand Pouillon, 944 Creteil Cedex, France, frederic.lafon@valeo.com Abstract Development of EMC test method to characterize IC and modelling technique such as ICEM afford new possibilities to perform design at system level. Whereas test methods are mainly used to perform validation tests or comparison between devices, the authors want to highlight the possibilities offered in term of modelling and simulation analysis.. INTRODUCTION In automotive industry, EMC specifications defined by carmakers can have very different requirements, even for the same tests in similar conditions. As an example we can note that more than 4 db difference can be required for a simple conducted emission test according CISPR5 standard []. The same tendencies can also be observed for immunity tests, such as the Bulk Current Injection (BCI), for which requirements can vary from Watts to Watts in term of incident power that can be applied on the injection probe []. From the need we have at car or equipment level, and considering this simple observation we can estimate the difficulty to identify the corresponding requirements at IC level using the general IC test methods proposed by IEC standards [3] [4]. This link can be difficult to establish, and is mainly based on experience. The EMC requirements at IC level will depend on the customer's objectives and specificity of the product such as the location inside the vehicle, the filters allowed and the PCB design technique used. This top down approach that consists in specifying an IC by considering always the worst cases seems unrealistic since it would conduct to very hard constraints not corresponding to the real need. We propose in this paper to show another approach which consists in using a bottom up approach, based on a systematic modelling technique of ICs for emission and immunity aspects. Some recent works with this approach already gives good results for specific case studied [5] [6]. This approach is also based on the used of a specification for ICs [7] but which objective is not a validation of the IC, but to provide all the information necessary to perform our simulation.. IMMUNITY APPROACH. DPI characterization The starting point of our analysis is based on the DPI characterization of an IC according the IEC 63 part 4 techniques [8]. The methodology will consist to estimate the differential voltage level induced on the input of a device that causes a failure to build our EMC model for immunity. This corresponds to the work introduced in [9]. If we consider the example of a hall sensor having 3 pins connection (+BAT / GND / output) we can establish the susceptibility behavior expressed in term of incident power for each pin of the device. On complex ICs the choice of the pin will depend on their accessibility in the final application. Pi ( dbm ) 5 45 4 35 3 5 5 5 Maximum injected level F ( MHz ) Fig.. DPI Results for incident power on +BAT pin of the hall sensor. Defect observed on output signal (±.5 V and ±.5µs tolerances) From this result we need to establish the correlation we have between the incident power and the differential induced voltage at the input of the IC. This is done by modeling the complete DPI set-up used to perform the test. The set-up for the DPI characterization of the device is given below:
SMA connector monitored signal To LISN and Power supply Hall sensor under test. 6. 5 3.3k SMA connector Amplifier output k Impedance (Ohms). 4. 3 DC Block ( nf // nf) High impedance Network (ferrites) Rotary motor Fig.. To LISN and Power supply Test set-up for DPI characterization of the hall sensor. The modeling of each part is done from measurement using a Network Analyzer (VNA), and from the results obtained equivalent models are built: a- Hall sensor modeling. The Hall sensor is modeled using the same methodology as for an ICEM Model []. We suppose that the impedance of this IC is not modified by the disturbance level we inject and that no significant non linearity appears. This can be checked during the DPI by measuring the input HF voltage induced on the pin and by comparing this result to the simulation considering a linear model (by also taking into account the oscilloscope input impedance in the simulation). Based on our experience on CAN, LIN bus and voltage regulators, this observation is verified but need to be checked for any new device. We propose an equivalent Passive Distribution Network (PDN) to model the IC, and measurement configurations are defined to extract all the parameters of the model. The PDN for the sensor is proposed on Fig. 3. Fig. 3. sensor Vbat Z Z GND Z3 Output Equivalent impedance structure model of the hall For example S measurement between the Vbat pin and the ground allows having Z in parallel with Z+Z3. An example of the impedance measured with correlation with the simulation is given below: Fig. 4.. 5. 6. 7. 8. 9 F(MHz) Measurement simulation Z impedance measurement compared to simulation. The final model obtained for the sensor is the following and is validated until GHz. Z R9 3e3 C5 e- R7 L5 C4 9e- R8 5 L6 3e-9 C3 5e- Output Sortie Z3 Z Alimentation + Bat Ground Masse Fig. 5. Equivalent model of the hall sensor extracted by measurement We can note that the model established by this method is similar to the one developed in [] for a voltage regulator application. The interest of such model is that is will allow to have correct impedance on the pin and that it will also allow to take into account the propagation of the disturbance through the IC (with the same linearity hypothesis as explained previously). It will allow for example to perform analysis at system level where other IC or microcontroller may be used to read the output information of this sensor. b- PCB modeling The various components used for injection (capacitances) and those functionally required need to be modeled integrating their parasitic elements having an influence
on the frequency range of interest (Until GHz). These models are established outside of the test PCB by S measurement with the network analyzer. The traces will be simply modeled using transmission line equivalent model. Its impedance depends on the PCB structure. In the PCB studied the traces are 6 Ohms impedance. The time delay of this line is calculated considering that the propagation speed is.4 m.s - since the PCB is FR4 made (ε r = 4.4). Length( m) TD = () 8.4 Some skin effect and radiation losses should also be considered. In a first approximation (overestimating for next analysis) we only consider skin effect losses on the PCB trace by using a frequency dependant model such as fig.6., or a model allowing time domain simulation such as []. TD = {Length/.4e8} Z = 6 Vac Vdc V R4 5 Amplifier.5uH e-6 e-3 C e-9 C L e-9 e-9 L3 e-9 R5 5e-3 R6 5e-3 RF coupling capacitors 5V Power supply Fig. 8..e- 3.3k + - H e-6 OUT+ IN+ OUT- IN- EFREQ OUT+ IN+ OUT- IN- EFREQ 3.3k resistance (,.,) (k,.,)(e9,.,) e6 PCB trace 5e- TD = {Length/.4e8} Z = 6 k resistance Oscilloscope input.e- Hall sensor model 5e- 3e-9 5 9e- Simulation model for DPI injection on +BAT This model allows establishing the transfer function between the level at the source and the level induced at the input of the device for V source: k V+ 3e3 V- e- e-6 + - H (,.,) (k,.,)(e9,.,) Fig. 6. Equivalent model for PCB traces taking into account skin effect. The comparison between measurement and simulation for the trace model is given below: V induced DPI - Spice.5. 6. 7 F(Hz). 8. 9 Z (Ohms). 5. 4. 3 Fig. 9. Simulation result for the differential voltage induced between +BAT and ground pins of the hall sensor for V source for the DPI generator. Since we know the real injection level to apply to produce a defect, expressed in term of incident power, it is possible to mix both information to estimate the final differential voltage on the device... 5. 6. 7. 8. 9 F(Hz) Measurement Simulation Fig. 7. Comparison between measurement and simulation for the trace model Vlimit (Volts) c- Amplifier injection The amplifier can be considered as a voltage source, 5 Ohms matched on its output. The cable connecting the output of the amplifier is not considered since also matched 5 Ohms. The voltage source is initially fixed at V... 6. 7. 8. 9 F(Hz) Fig.. Maximum differential voltage on +BAT before having a defect on the hall sensor. We finally have for the IC its equivalent impedance model and the maximum acceptable differential voltage
that produce a defect. That's all the necessary information to perform the analysis at the system level. To be able to produce such model from DPI characterization it implies to have a good knowledge about the test set-up used so that it can be modeled. With this technique the maximum level allowed will correspond either to the level for which the device is disturbed or to the maximum level for which the device had been tested without defect. We won't distinguish the both status and we can see the importance to perform the characterization at high injection level even if it doesn't correspond to a validation level. The management of these two points is dealt through the IC specification [7].. System analysis The previous model can be now used in a complete system and we can perform a risk analysis allowing defining the needs in term of filtering and layout constraints. We can for example estimate the disturbance level induced on the input of the device when applying BCI on a PCB on which the hall sensor is used. The method used will consist in modeling all the parts involved in the BCI test bench and that will be the same approach for any system test of interest: LISN/ Loads Injection probe ma Fig.. BCI General test set-up. Device under test is connected on the right. a- LISN and loads The LISN used in the test set-up and the loads have to be modeled to take into account their parasitic elements on the frequency range of interest. These models can be extracted by S measurement with a network analyzer. The modeling technique is similar to the ICEM modeling technique. The LISN model is not of interest here and is similar to the one produce in [3] b- Harness Harness is modeled considering coupled transmission lines. The length and height are specified in the standard or in the carmakers specifications. The particularity is that automotive harnesses are in fact non uniform transmission lines, since the location of each wire regarding the other one can vary along the bundle. For our current analysis we will consider uniform model but we will have to consider the dispersion that could occur due to this parameter in the final risk estimation. Tendencies shows that dispersions about db can be observed due to this parameter [6] [3] [4]. c- Injection probe A lumped elements model of the injection probe can be used for our analysis. This kind of model had been developed first in [5] and more recently in [6]. The modeling of the disturbance source is one of the key points of the system analysis and the accuracy of this model is very important. d- Device under test The device under test includes the IC (hall sensor), but also other ICs, passive components and traces linking these different devices. The modeling techniques used are similar to the ones describe in the previous part. The last parameter that needs to be estimated is the common mode impedance between the PCB and the reference ground plane. This one is due to particular configuration where the PCB is floating and doesn't have direct connection to the chassis. This will be modeled using simple capacitor effect [6]. The final model is given on fig.. PCB model T4 T5 Hall sensor model TD = {.3/.4e8} TD = {./.4e8} BAT DUT VAMP = 4 Z = 6 L e-9 C6 e-9 Z = 6 V+ 5e- 3e-9 LISN_R_ext U5 R_ R_ 5 U MTL Model Wires cm GHz VAMP = 4 U MTL Model Wires 9 cm GHz R 5e-3 5 3e3 e- reference reference V- 9e- VBAT GND BAT DUT R_ BATTERIEV LISN_R_ext U6 R_ 5.5e- Micro controller input C3 4e- T3 TD = {.3/.4e8} Z = 6 k L9 C4 7e- R9 e6 Fig.. BCI Model for system analysis
This kind of approach and modeling technique had been applied in [6] where correlation between measurements and simulation give good results. This allows validating these modeling techniques. At system level comparison between differential level induced on the hall sensor and the maximum acceptable level established Fig. can be done. V (Volts).. 3. 4.. 6. 7 F(Hz). 8. 9 Susceptibility limit Induced voltage for BCI - no filter Induced voltage for BCI - nf filter Fig. 3. Induced level on the sensor during BCI test system. This allows estimating the risk to disturb this function of the product. We can note Fig.3 in particular that without filters the induced level is higher than the susceptibility level. This means that filters need to be used on this function. With the example of nf capacitor added we can see that the level is globally below the objective. But this can be not enough. In this simulation the difference between the induced level and the limit can allow to estimate a probability to disturb the function, which should be the only acceptance criteria to validate the result. We can apply here a resistance constraint approach, widely used in mechanic domain. Depending on complexity of repartitions' law the failure probability can be computed manually or require numerical tools. We will limit here to a simple case where constraint and resistance are Gaussian's law. The resistance level Vlimit established on Fig.. will be noted µr. On this level we can associate dispersion σr due to DPI test set-up and more especially to the power step used for the tests. In our case 3.σr= 5 db. For the constraint, we can note µc the level induced using the system test. To this test we can estimate that some dispersions will be caused by the harness structure (as discussed previously), but also more generally to all elements of the BCI test set-up (probes' locations, LISN characteristics, distance from faraday cage hall ). Based on experience [3] and [4] we can estimate that dispersion about db can be expected (3.σc). When the constraint and the resistance are Gaussian, then the resistance is also Gaussian with µz and σz given by: µ z = µ r µ c () σ z + = σr σc () And the repartition function is given by: t µz exp σz Fz ( t) = (3) σz π with t: constraint level. We finally obtain the failure probability with: Failure probability. 4. 5. 6. 7. 8. 9.... 4 π.. 3 µz x σz exp. dx F = (4) π. 6. 7. 8. 9 F (Hz) Fig. 4. Failure probability during BCI tests with nf filter on the sensor At car level the same approach can be used. In addition for the risk estimation the interaction probability between the disturbance and the system must also be took into account. 3. EMISSION APPROACH 3. Model extraction from Ohm measurement The analysis for conducted emissions will be based on the use of the ICEM model. This model is basically composed by a Power Network Model (PDN) and a noise source called Internal Activity (IA). The extraction of the PDN is performed using the measurement techniques described in the previous part. The PDN extracted for immunity will be so the same than for emissions. The second part of the model will consist in the current source. Its extraction can be done using the Ohm and/or 5 Ohms conducted emission measurement methods [7]. The equivalent model of the test set-up can be built with the same approach and techniques used in the previous part.
3. System analysis The use of an ICEM model at system level had already made the object of several works such as [8] and [9]. It allows having good correlation between measurement and simulation, such as for the case below, for a SX microcontroller which model had been developed by []. U (dbµv) 5 4 3. 7 4. 7 6. 7 8. 7. 8.. 8.4. 8.6. 8.8. 8. 8.. 8 F (Hz) Mesure Simulation Fig. 5. Conducted emissions simulation compared to measurement for system level. As for immunity the accuracy of the result will depend on the correct modelling of the test set-up used to perform this measurement. The specification [7] allows managing this point. 4. CONCLUSION We have shown in this paper how results obtained using IC test methods can be used to build EMC models for immunity and emission analysis at higher level. The accuracy of the models is linked to the management of the test methods and influent parameters for the experimental characterization. The aim of the Valeo specification [7] is to manage these points by defining tests methods parameters important to manage for the tests. At each integration level the dispersion sources are more and more important and the authors wanted to highlight that the simulation and risk analysis should be based on a statistical approach. An introduction of such approach had been described in this paper. [3] IEC 6967 Integrated circuits Measurement of electromagnetic emissions 5 khz to GHz: Part : General conditions and definitions [4] IEC 63 Integrated circuits Measurement of electromagnetic immunity 5 khz to GHz - Part : General conditions and definitions [5] R. Neumayer, R. Weigel, A. Stelzer, G. Steinmair, F. Haslinger, M. Tröscher, J. Held, B. Unger : Numerical EMC Simulation for Automotive Applications - 5th International Zurich Symposium on EMC, Feb 3. [6] S.Egot and all, "Modeling Automotive Electronic Equipment in a Realistic Sub-system", in proceedings of EMC Europe 6, Barcelone, pp4-45 [7] F.Lafon, "VALEO EMC Specification for ICs 5-449- 96E" [8] IEC 63 Integrated circuits Measurement of electromagnetic immunity 5 khz to GHz Part 4: Direct RF Power Injection to measure the immunity against conducted RF disturbances of integrated circuits up to GHz. [9] F.Lafon and All, "ICEM-ICIM Modeling and exploitation for bus transceivers applications" EMC COMPO Proceedings -4 [] IEC6433- Models of Integrated Circuits for EMI behavioral simulation - ICEM-CE, ICEM Conducted Emission Model [] F.Fiori "EMC Issues in Linear Voltage regulator Circuit Design for SoC Applications" EMC COMPO Proceedings 5 Munich [] Chu-Sun Yen "Time-Domain skin effet model for transient analysis of lossy transmission lines". IEEE transactions on Electromagnetic compatibility, vol. 7, no 7, pp75-757, July 98 [3] F.Lafon, "Harness influence in Bulk Current Injection testing". EMC Zurich Symposium 5 - CD proceedings. [4] S.Egot and al. "Influence of the PCB traces of an automotive electronic equipment in the case of random cable harnesses" EMC Zurich Symposium 5 CD proceedings. [5] F. Duval, "Bulk Current Injection Test Modeling and creation of a test methodology ", EMC Zurich 3, pp.493-498 [6] F. Grassi "Improved Lumped-Pi Circuit Model for Bulk Current Injection Probes" Proceedings of IEEE Symposium on Electromagnetic Compatibility 5. [7] IEC 6967 Integrated circuits Measurement of electromagnetic emissions, 5 khz to GHz Part 4: Measurement of conducted emissions Ω/5 Ω direct coupling method. [8] F.Lafon and Al. "Exploitation of the ICEM model in an automotive application", EMC COMPO 4 Angers CD Proceedings -4 [9] F.lafon and Al. "Analyse de risque au niveau système par l'exploitation du modèle ICEM" EMC6, Saint Malo, CD Proceeding. [] S. Ben Dhia, "6-bits microcontroller emission measurements and modeling using different approaches", EMC COMPO 4 Angers CD Proceedings -4 5. REFERENCES [] CISPR5, "Radio disturbance characteristics for the protection of receivers used on board vehicles, boats and devices Limits and methods of measurements ", -8 [] ISO 453-4 Standard " Road vehicles Component test methods for electrical disturbances from narrowband radiated electromagnetic energy Part 4: Bulk current injection (BCI)"