TWI761177B - Method for flattening impedance of power deliver network - Google Patents

Abstract

A method for flattening an impedance of a power deliver network, includes capturing a set of impedance parameters; obtaining an impedance of the power deliver network according to the set of impedance parameters; defining a target impedance; performing an importance calculation to determine a port, obtaining an intersection frequency according to the target impedance and the impedance of the power deliver network; selecting a decoupling capacitor according to the intersection frequency; and disposing the decoupling capacitor on the port. Hence, the method of the invention provides the effect of reducing the impedance of the power deliver network to the target impedance and flattening the impedance, so the rogue wave phenomenon can be avoided.

Description

Method for Flattening the Impedance of Power Delivery Networks

The present invention relates to a method for flattening the impedance of a power delivery network, and more particularly, to a method for selecting a suitable decoupling capacitor to flatten the impedance of the power delivery network.

A power distribution network (PDN) provides a transmission medium for a voltage source to transmit DC power to a load, and is usually a transmission path from a voltage regulator module (VRM) to a load circuit on a printed circuit board. The power delivery network can be composed of the power supply layer and the ground layer of the printed circuit board, and components such as cables, connectors, and capacitors that are electrically connected to the power supply layer and the ground layer can be part of the power delivery network.

Since parasitics of the power delivery network can cause parallel resonance, high self-impedance at the location of the load circuit results. At the position of the load circuit, the problem of voltage disturbance will arise due to the variation of the load current and the high self-impedance of the power delivery network.

In addition, when the self-impedance versus frequency curve of the power delivery network is not flat, the problem of rouge waves caused by overlapping voltage disturbances will make the voltage disturbance problem even more serious. A server is an electronic device including a printed circuit board. The power supply layer and the ground layer of the printed circuit board can form a part of the power delivery network. Therefore, the server needs to be considered in the design of the power delivery network of the printed circuit board. In addition to reducing the impedance to avoid noise interference, it is also necessary to consider the issue of impedance flattening to avoid distortion.

An embodiment provides a method for flattening the impedance of a power delivery network, including acquiring a set of impedance parameters of the power delivery network; obtaining the impedance of the power delivery network according to the set of impedance parameters; defining a target impedance; according to the target impedance and the impedance of the power delivery network, obtain the intersection frequency; according to the intersection frequency, select a decoupling capacitor; and set the decoupling capacitor at the port.

100: Method

110 to 199, 310, 320, 410 to 450, 510 to 530: Steps

Z11, Z11', Z11'a, Z11'b, Z11'c: Impedance

Fh: predetermined frequency

Fx, Fx', Fx1, Fx2, Fx3: Intersection frequency

FIG. 1 is a flowchart of a method for flattening the impedance of a power delivery network in an embodiment.

Fig. 2 is a graph showing the relationship between impedance and frequency in the embodiment.

Figures 3 to 5 are the flow charts of the selection of decoupling capacitors in Figure 1.

FIG. 6 is a graph of executing the process of FIG. 5 .

To address the aforementioned challenges, decoupling capacitors can be placed to reduce the self-impedance of the power delivery network. However, although the impedance of the power delivery network can be reduced to the target impedance, the impedance of the power delivery network has not been flattened and optimized. Embodiments provide methods for planarizing and optimizing the impedance of a power delivery network, as described below.

FIG. 1 is a flow diagram of a method 100 of planarizing the impedance of a power delivery network, in accordance with an embodiment. Fig. 2 is a graph showing the relationship between impedance and frequency in the embodiment. As shown in FIG. 1 and FIG. 2 , the method 100 may include the following steps: Step 110 : acquiring a set of impedance parameters of the power delivery network; Step 120 : generating the impedance Z11 of the power delivery network according to the set of impedance parameters; Step 130: Define the target impedance Ztarget; Step 140: Perform importance calculation to determine the port position; Step 150: Obtain the intersection frequency Fx according to the target impedance Ztarget and the impedance Z11 of the power delivery network; Step 160: According to the intersection frequency Fx, select the decoupling capacitor; Step 170: Set the decoupling capacitor at the port; Step 180: Perform a parallel operation to generate the adjusted impedance Z11' of the power delivery network; Step 190: Observe whether the adjusted impedance Z11' meets the target impedance Ztarget; if yes, go to step 199; if no, go to step 140; step 199: end.

The steps of FIG. 1 are described below by way of example. In step 110 , the scattering parameters (S-parameters) of the power delivery network can be acquired, and the impedance parameters can be obtained by calculation according to the scattering parameters, wherein the impedance parameters can be the impedance matrix of the power delivery network (impedance matrix, also known as Z-matrix) elements (elements).

In step 120, the impedance Z11 of the power delivery network can be obtained according to the set of impedance parameters. The curve of impedance Z11 is shown in FIG. 2, and the curve of impedance Z11 can represent the self-impedance of the power delivery network at each frequency at this time. FIG. 2 is only an example, rather than limiting the aspect of the embodiment. In Figure 2, 1e+05 can represent 1×10 ⁵ , 1e+06 can represent 1×10 ⁶ , and so on.

The power delivery network can include multiple ports, the impedance Z11 can be the input impedance of the port coupled to the load device (such as an integrated circuit), and other ports can be used to set decoupling capacitors to flatten the impedance Z11, as described below.

In step 130, the target impedance Ztarget can be defined according to the actual requirements and product specifications. As shown in FIG. 2, the curve of the target impedance Ztarget can correspond to the target impedance value of each frequency. In the frequency band, the target impedance Ztarget can be substantially constant. Step 130 may be independent of Steps 110 and 120, for example, step 130 can also be performed first, and then steps 110 and 120 can be performed.

According to an embodiment, as described in steps 140 to 170, the most important port may be selected from a plurality of ports in the power delivery network, and then the most suitable capacitor may be selected as the decoupling capacitor among the plurality of candidate capacitors. to set at the selected port.

In step 140, a plurality of inductance characteristic values that are reduced after a plurality of ports of the power delivery network are set with capacitors can be calculated respectively, and the port corresponding to the highest one of the plurality of inductance characteristic values is selected as the selected one. port. In other words, among multiple ports, the port whose inductance characteristic decreases the most after the capacitors are set can be the most important port.

其中，L11及L22可為第一埠位及第二埠位之自感，L12可為第一埠位及第二埠位之互感，L_11wdecap@2可為第二埠位被設置去耦電容後的電感矩陣之調整後元素，且△L_11wGND@2可為第二埠位被設置去耦電容後所降低之電感特性。其中，Ld可為所設置的去耦電容之自感值，可為常數，為了計算方便，可予以忽略。上述算式eq-1、eq-2僅為舉例，若於n-埠(n-port)網路，則相關於第A埠之互感的通式可如下述：

舉例而言，若電源遞送網路為10埠(10-port)網路，則可根據算式eq-2之原理，求得表1：

For example, if the power delivery network is a 2-port network, it has a first port (eg, coupled to a load device, such as an integrated circuit) and a second port (eg, coupled to a decoupling) Capacitance), then the calculation of formula eq-1 and eq-2 can be performed:

Wherein, L11 and L22 can be the self-inductance of the first port and the second port, L12 can be the mutual inductance of the first port and the second port, and _L11wdecap@2 can be the decoupling capacitor set for the second port The adjusted element of the latter inductance matrix, and ΔL _11wGND@2 can be the inductance characteristic reduced after the second port is set with a decoupling capacitor. Among them, Ld can be the self-inductance value of the set decoupling capacitor, which can be a constant, and can be ignored for the convenience of calculation. The above formulas eq-1 and eq-2 are only examples. For an n-port network, the general formula related to the mutual inductance of the A-th port can be as follows:

For example, if the power delivery network is a 10-port network, Table 1 can be obtained according to the principle of the formula eq-2:

In the example in Table 1, it can be seen that when the decoupling capacitor is set at the port P2, the inductance characteristic decreases the most, so the port P2 is the most important, and the port P2 can be selected in step 140. The number and value of the ports in Table 1 are only examples, and the embodiment is not limited thereto.

In step 150, as shown in FIG. 2, the intersection frequency Fx can be obtained on the impedance versus frequency curve according to the curve of the target impedance Ztarget and the curve of the self-impedance Z11 of the power delivery network at this time.

In step 160 , a decoupling capacitor can be selected from the decoupling capacitor library, and then in step 170 the selected decoupling capacitor is set to the port determined in step 140 .

After step 170 is executed, each element of the equivalent impedance matrix of the power delivery network can be changed due to the setting of the decoupling capacitor, so the self-impedance of the power delivery network needs to be recalculated.

For example, if the power delivery network has 100 ports and one port coupled to a load device (such as an integrated circuit), the dimension of the original impedance matrix can be 101×101, and step 170 is executed to set one After decoupling capacitors, the dimension of the impedance matrix can be 100×100, and each element of the matrix may be changed, so the impedance matrix must be regenerated and the self-impedance calculated accordingly.

n。Zd可為所設置的去耦電容之阻抗。 In step 180, a parallel operation may be performed according to the decoupling capacitor set in step 170 to generate a curve of the adjusted impedance Z11' of the power delivery network. The parallel operation can be as shown in the formula eq-5: Z11'=Z11-(Z1i×Zi1)/Z11+Zd...eq-5; wherein Z11 can be the part of the power delivery network before the adjustment. Self-impedance, Z11' can be the self-impedance of the adjusted power delivery network after setting the decoupling capacitor, Z1i and Zi1 can be the elements of the impedance matrix, if the matrix dimension is n×n, then i and n are positive integers and 0<i

n. Zd can be the impedance of the set decoupling capacitor.

In step 190, it can be observed whether the adjusted impedance Z11' conforms to the target impedance Ztarget. If so, step 199 can be entered to end the process; Set an appropriate decoupling capacitor at the selected port, so that the impedance curve of the power delivery network is closer to the target impedance Ztarget curve on the impedance-frequency coordinate, and is flatter.

In step 190, whether the adjusted self-impedance Z11' meets the target impedance Ztarget can be determined at least according to whether the adjusted impedance Z11' is smaller than the target impedance Ztarget in a specific frequency band. In addition, it can be judged whether the flatness is good enough according to the standard deviation and average value of the adjusted impedance Z11' in a specific frequency band. For example, the standard deviation and average value of the adjusted impedance Z11' in a specific frequency band can be divided and expressed as a percentage to obtain the flatness.

Regarding steps 140 to 190, for example, in step 140, a first importance calculation is performed to determine the first port, and as shown in FIG. 1, steps 150 to 180 are performed to set the first decoupling capacitor on the first port. a port and obtain the adjusted impedance; if in step 190, the adjusted impedance Z11' does not yet meet the target impedance Ztarget, step 140 can be performed again to perform the second importance calculation to determine the second port, and then the execution Steps 150 to 180 are performed to set the second decoupling capacitor at the second port and obtain the adjusted impedance; in this way, the flatness of the adjusted impedance Z11' can be improved by a loop method. The first importance calculation and the second importance calculation mentioned here are only examples, and are used to describe that the steps in FIG. 1 can be executed in a loop manner. However, the embodiment is not limited to this. The steps may perform more (eg, more than two) importance calculations in a loop fashion, thereby improving the flatness of the adjusted impedance Z11'. In this example, as described above, the first importance calculation may include separately calculating a plurality of inductance characteristic values reduced after a plurality of ports of the power delivery network are set with capacitances, and selecting the highest one of the plurality of inductance characteristic values The corresponding port is regarded as the first port; similarly, the second importance calculation may include calculating the plurality of inductance characteristic values that are reduced after the capacitances are set on the plurality of ports of the power delivery network, and selecting the plurality of inductance characteristics. The port corresponding to the highest inductance characteristic value is used as the second port; if there is a third importance calculation, a fourth importance calculation, etc., it can be deduced by analogy.

FIG. 3 is a flowchart of selecting a decoupling capacitor in step 160 of FIG. 1 . According to an embodiment, step 160 in FIG. 1 may include the following steps: Step 310 : Select a set of capacitors, and the self-resonant frequency (SRF) of each capacitor in the set of capacitors is between the first value and the between the second values; Step 320: Select one of the capacitors as a decoupling capacitor.

The first value mentioned in Figure 3 may be A times the intersection frequency Fx, and the second value may be B times the intersection frequency Fx, A and B are positive parameters, and 0<A<B.

For example, if there are 200 capacitors for selection in the database, and the self-resonant frequency of 30 capacitors is between A×Fx and B×Fx, one of the 30 capacitors can be selected as the decoupling capacitor. According to the embodiment, 0.5<A<B<5. According to an embodiment, A may be substantially 1.5 and B may be substantially 3.

FIG. 4 is a flowchart of selecting a decoupling capacitor in step 160 of FIG. 1 . According to the embodiment, Step 160 in FIG. 1 may include the following steps: Step 410 : select k capacitors; Step 420 : calculate the parallel value of each of the k capacitors and the self-impedance of the power delivery network respectively, so as to obtain k parallel connections Impedance value; Step 430: Is there a parallel impedance value within a specific range among the k parallel impedance values? If yes, go to step 440; if no, go to step 460; Step 440: Select m parallel impedance values within a specific range among the k parallel impedance values; Step 450: Select the m capacitors corresponding to the m parallel impedance values One is a decoupling capacitor; and Step 460 : Adjust the specific range, and proceed to Step 430 .

k。步驟430之特定範圍介於目標阻抗Ztarget之α倍至β倍之間，且0<α<β。 In Figure 4, m and k are positive integers, and 0<m

k. The specific range of step 430 is between α times and β times the target impedance Ztarget, and 0<α<β.

For example, if 100 candidate capacitors are respectively set at the port positions determined in step 140 in FIG. 1, 100 parallel impedance values can be obtained, that is, the adjusted self-impedance Z11' of 100 power delivery networks. Among the 100 adjusted self-impedances Z11', if there are 40 adjusted self-impedances that can fall within the range of α×Ztarget to β×Ztarget, then the 40 candidate capacitors corresponding to the 40 adjusted self-impedances can be selected. One is the decoupling capacitor, and the remaining 60 candidate capacitors are not selected.

For example, α can be 0.9, and β can be 1. If in step 430, no parallel impedance value is within a specific range, α can be decreased to increase the specific range described in step 430, so that the Step 430 selects a shunt impedance value within a specific range. For example, α can be gradually adjusted from 0.9 to 0.85, 0.8, 0.75, etc. to gradually increase the specific range. According to another embodiment, α and β can also be adjusted so that the specific range after the adjustment in step 460 is different from the specific range before the adjustment.

FIG. 5 is a flowchart of selecting a decoupling capacitor in step 160 of FIG. 1 . According to the embodiment, Step 160 in FIG. 1 may include the following steps: Step 510: Calculate the parallel value of each capacitor of the plurality of capacitors and the self-impedance Z11 of the power delivery network to obtain a plurality of reference impedance curves; Step 520: According to A plurality of reference impedance curves and a target impedance are used to generate a plurality of reference intersection frequencies; and Step 530 : Select a capacitor corresponding to the highest of the plurality of reference intersection frequencies as a decoupling capacitor.

FIG. 6 is a graph of executing the process of FIG. 5 . FIG. 6 is only an example, rather than limiting the aspect of the embodiment. For example, after the three candidate capacitors are respectively set at the ports determined in step 140 in FIG. 1, three reference impedances Z11'a, Z11'b and Z11'c and their reference impedance curves can be obtained. As shown in FIG. 6 , according to the curve of the target impedance Ztarget and the reference impedance curve, the reference intersection frequencies Fx1 , Fx2 and Fx3 can be obtained, wherein the reference intersection frequency Fx3 is the highest of the three reference intersection frequencies. Therefore, among the three candidate capacitors, the capacitor corresponding to the curve of the reference intersection frequency Fx3 and the reference impedance Z11'c can be selected as the decoupling capacitor.

The process shown in Figure 3, Figure 4, and Figure 5 can be used to select decoupling capacitors step by step from the capacitor database. For example, if there are 100 candidate capacitors, you can first select 30 capacitors using the process in Figure 3, then select 10 capacitors from the 30 capacitors using the process in Figure 4, and then use the process in Figure 5 to select 10 capacitors. One capacitor is selected from the ten capacitors as the decoupling capacitor selected in step 160 of FIG. 1 . However, this is only an example, and the sequence of executing the processes in FIG. 3 , FIG. 4 , and FIG. 5 is not limited thereto.

In conclusion, with the method provided by the embodiments, the most important port position can be effectively selected, and an appropriate decoupling capacitor can be selected to be set at the selected port position. With the loop method shown in Figure 1, after setting a suitable decoupling capacitor at a suitable port position, it can help to flatten the self-impedance of the power delivery network. According to experiments, the curve of the adjusted impedance Z11' on the impedance-frequency coordinate obtained by the method of the embodiment can be closer to the curve of the target impedance Ztarget, that is, has better flatness. Therefore, the method of the embodiment can be effective Avoid distortion. Therefore, it is helpful to deal with the problems in this field.

The method of the present invention can be applied to the design of the printed circuit board of the server, so that the server has a high-quality power delivery network, so as to improve the stability of the server, so that the server can be applied to artificial intelligence (Artificial Intelligence, referred to for short). AI) computing, edge computing (Edge Computing), can also be used as 5G server, cloud server or car networking server.

The above descriptions are only preferred embodiments of the present invention, and all equivalent changes and modifications made according to the scope of the patent application of the present invention shall fall within the scope of the present invention.

100: Method

110 to 199: Steps

Claims (10)

A method for flattening the impedance of a power delivery network, comprising: acquiring a set of impedance parameters of the power delivery network; obtaining an impedance of the power delivery network according to the set of impedance parameters; defining a target impedance; a first importance calculation to determine a first port; obtain an intersection frequency according to the target impedance and the impedance of the power delivery network; select a first decoupling capacitor according to the intersection frequency; and set The first decoupling capacitor is at the first port. The method of claim 1, further comprising: performing a parallel operation to generate an adjusted impedance of the power delivery network; performing a second importance calculation to determine a second port; according to the target impedance and The adjusted impedance of the power delivery network is to obtain an adjusted cross-point frequency; according to the adjusted cross-point frequency, a second decoupling capacitor is selected; and the second decoupling capacitor is set at the second port. The method of claim 1, wherein selecting the first decoupling capacitor comprises: selecting a set of capacitors, wherein a self-resonant frequency of each capacitor in the set of capacitors is between a first value and a second value and selecting one of the capacitors as the first decoupling capacitor; wherein the first value is A times the frequency of the intersection, the second value is B times the frequency of the intersection, A and B are positive parameters, And 0<A<B. The method of claim 3, wherein A is substantially 1.5. The method of claim 3, wherein B is substantially 3. The method of claim 1, wherein selecting the first decoupling capacitor comprises: selecting k capacitors; separately calculating a parallel value of each of the k capacitors and a self-impedance of the power delivery network, to obtain Obtain k shunt impedance values; select m shunt impedance values within a specific range among the k shunt impedance values; and select one of m capacitors corresponding to the m shunt impedance values as the first Coupling capacitor; where m and k are positive integers, and 0<m

k. The method of claim 6, wherein the specific range is between α times and β times the target impedance, and 0<α<β. The method of claim 1, wherein selecting the first decoupling capacitor comprises: selecting k capacitors; separately calculating a parallel value of each of the k capacitors and a self-impedance of the power delivery network, to obtain Obtaining k parallel impedance values; selecting a parallel impedance value within a first specific range among the k parallel impedance values; and selecting the k parallel impedance values when none of the k parallel impedance values falls within the first specific range Among the shunt impedance values, m shunt impedance values within a second specific range; and selecting one of m capacitors corresponding to the m shunt impedance values to be the first decoupling capacitor; wherein m and k are positive Integer, 0<m

k. The method of claim 1, wherein selecting the first decoupling capacitor comprises: calculating a parallel value of each capacitance of a plurality of capacitors and a self-impedance of the power delivery network respectively to obtain a plurality of reference impedances; generating a plurality of reference intersection frequencies according to the plurality of reference impedances and the target impedance; and selecting A capacitor corresponding to the highest one of the plurality of reference intersection frequencies is the first decoupling capacitor. The method of claim 1, wherein performing the first importance calculation comprises: separately calculating a plurality of inductance characteristic values reduced after a plurality of ports of the power delivery network are set with capacitance; and selecting the plurality of inductance characteristic values A port corresponding to the highest inductance characteristic value is used as the first port.

Images