Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
This action is in response to the application and claims filed 03/07/2024. Claims 1-11 are
pending and have been examined. Claims 1-11 are rejected.
Claim Rejections - 35 USC § 112
The following is a quotation of 35 U.S.C. 112(b):
(b) CONCLUSION.—The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the inventor or a joint inventor regards as the invention.
The following is a quotation of 35 U.S.C. 112 (pre-AIA ), second paragraph:
The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the applicant regards as his invention.
Claim 2-3, 4-5 are rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor (or for applications subject to pre-AIA 35 U.S.C. 112, the applicant), regards as the invention.
Regarding claim 2, the recited group includes a decision tree algorithm, a gradient boosting regression algorithm, and a k-nearest neighbor algorithm. None of these from the group trains "a network weight and a bias" as required by claim 1. When one of these members is selected, it cannot be determined how the training step of claim 1 is performed, rendering the claim scope unclear. Claim 3 is rejected as being dependent on claim 2.
Claim 4 recites the limitation "the bottom cell". There is insufficient antecedent basis for this limitation in the claim. Because earlier in the claim it recites "at least one bottom cell," it cannot be determined when a plurality of bottom cells is present, above which bottom cell(s) the sub-cells must be disposed. Claim 5 also inherits similar issues regarding it’s recitation of “the bottom cell” and “tunnel junction of the bottom cell”
The term "upper" in claim 5 is a relative term which renders the claim indefinite. The term "upper" is not defined by the claim, the specification does not provide a standard for ascertaining the requisite degree, and one of ordinary skill in the art would not be reasonably apprised of the scope of the invention. If there are three or more sub-cells, "an upper sub-cell" may be the uppermost sub-cell or any sub-cell above another, so the identity of "the remaining sub-cells" and which sub-cells carry tunnel junctions versus which carry the contact layer cannot be determined. For examination, "an upper sub-cell" is interpreted as an uppermost sub-cell consistent with [0029]–[0030] of the instant application.
Claim Objections
Claim 1 objected to because of the following informalities:
“optimizing the initialized model: training…” should read “optimizing the initialized model: by training…”
Appropriate correction is required.
Claim 7 objected to because of the following informalities:
“and to covert the processed feature data…” should read “and to convert the processed feature data”
Appropriate correction is required.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
This application currently names joint inventors. In considering patentability of the claims the examiner presumes that the subject matter of the various claims was commonly owned as of the effective filing date of the claimed invention(s) absent any evidence to the contrary. Applicant is advised of the obligation under 37 CFR 1.56 to point out the inventor and effective filing dates of each claim that was not commonly owned as of the effective filing date of the later invention in order for the examiner to consider the applicability of 35 U.S.C. 102(b)(2)(C) for any potential 35 U.S.C. 102(a)(2) prior art against the later invention.
Examiner’s Note: Some rejections will include an Examiner’s Note (labeled ‘EN’) to provide additional context or rationale explaining the basis for the rejection.
Claim 1-3, 9-11 is rejected under 35 U.S.C. 103 as being unpatentable over Yi et al., "Tandem Solar Cells Efficiency Prediction and Optimization via Deep Learning," hereinafter “Yi”, in view of Qiu et al. (US 2022/0122103 A1), hereinafter “Qiu”.
Regarding claim 1, Yi teaches:
"A method for predicting performance of solar cell structure, comprising the following steps:" (Yi, Abstract, p. 1: "a deep neural network is established to predict the achievable short-circuit current for tandem solar cells with given cell structure").
"collecting and extracting input feature parameters of solar cell structures," (Yi, Sec. 2.3, p. 3: "we formulate a data set with of 12,500 sets of numerical simulation data from FDTD, which are paired data of the set of layer thickness (in unit of nm) and current properties of tandem solar cells"; Sec. 2.3, p. 3: "layer thickness of tandem solar cells {Hg, He, Hm, Hp, Hi} as input data").
"output feature parameters corresponding to the input feature parameters," (Yi, Sec. 2.3, p. 3: "For each layer thickness set, we run the FDTD simulation of the tandem solar cell and obtain the corresponding MACD of top sub-cell It, MACD of bottom sub-cell Ib and reflection loss Ir"; Sec. 2.3, p. 3: "current properties {It, Ir, Ib} as output data").
"and input feature parameters of a to-be-predicted solar cell structure;" (Yi, Sec. 3.1.1, p. 4: "The data set formulated by FDTD is randomly divided into five subsets for five-fold cross validation while training, four of which are adopted as training data set and the other is evenly divided as validation and test data set"; Sec. 3.1.2, p. 4: "Fig. 7 further presents the comparison of the absolute values of It, Ir and Ib predicted by CPN and simulated by FDTD. Due to the limited space, here we randomly select 50 sets out of test data for the sake of illustration." - EN: under the broadest reasonable interpretation, the "to-be-predicted solar cell structure" reads on Yi's held-out test structures, whose layer-thickness sets constitute the input feature parameters forming the test data set and are input to CPN to obtain predicted current properties It, Ir, Ib.)
"and taking the input feature parameters of solar cell structures and the output feature parameters corresponding to the input feature parameters as a training data set, and taking the input feature parameters of the to-be-predicted solar cell structure as a test data set;" (Yi, Sec. 3.1.1, p. 4: "four of which are adopted as training data set and the other is evenly divided as validation and test data set", -- EN: the training data set containing the paired thickness inputs and current outputs, Sec. 2.3, p. 3).
"constructing an initial model by using a machine learning algorithm;" (Yi, Sec. 2.2, p. 3: "a deep fully-connected neural network, named current prediction network (CPN), is designed and trained to predict the current properties of tandem solar cells with given layer thickness, replacing the FDTD numerical simulation").
"setting structural parameters of the initial model," (Yi, Sec. 2.3, p. 3: "Fig. 4 presents the architecture of CPN, which is a 15-layer deep neural network"; Sec. 2.3, p. 3: "CPN follows a series-parallel-series structure to reduce the number of training parameters in network"; Fig. 4 caption, p. 4: "FC-X in architecture represents there are X hidden neurons in this fully connected layer").
"optimizing the initialized model: training the initialized model by using the ... training data set to obtain a network weight and a bias, thereby obtaining a prediction model;" (Yi, Sec. 2.2, p. 3: "a deep fully-connected neural network, named current prediction network (CPN), is designed and trained to predict the current properties of tandem solar cells"; Sec. 2.3, p. 3: "CPN follows a series-parallel-series structure to reduce the number of training parameters in network"; Sec. 3.1.1, p. 4: "four of which are adopted as training data set"; Sec. 3.1.1, p. 4: "the training process converges at about 600 epochs". – EN: Yi trains its fully-connected CPN on the training data set until the training converges, and the "training parameters" which are learned are a network weight and a bias, since each fully-connected layer computes a weighted sum of its inputs plus a bias term. The converged CPN is then the claimed prediction model, as Yi uses it to predict the current properties in place of the FDTD simulation. The preprocessed character of the training data set is addressed in the combination with Qiu below. On a side note, Qiu also computes weight and bias, see Qiu [0052].)
"inputting the ... test data set into the prediction model to obtain predicted values of output feature parameters of the to-be-predicted solar cell structure." (Yi, Sec. 3.1.2, p. 4: "Fig. 7 further presents the comparison of the absolute values of It, Ir and Ib predicted by CPN and simulated by FDTD. ... here we randomly select 50 sets out of test data"; – EN: the preprocessed character of the test data set is addressed in the combination with Qiu below).
Yi does not explicitly teach
"preprocessing the training data set and the test data set to obtain a preprocessed training data set and a preprocessed test data set",
"performing initialization training on the structural parameters to obtain an initialized model", or, correspondingly, the preprocessed character of the training data set and the test data set recited in the optimizing and inputting steps.
"preprocessing the training data set and the test data set to obtain a preprocessed training data set and a preprocessed test data set" (Qiu [0029]: "performing data preprocessing on the historical actual measurement data set and the calculation simulation data set, including data denoising, data supplement and data normalization processing"; Qiu Fig. 1: "Perform data preprocessing, and divide the calculation simulation data set into a training sample set, a verification sample set and a test sample set" – EN: Qiu's data denoising and normalization is the claimed preprocessing, and it operates on Qiu's two data sets, which are divided into a training sample set and a test sample set; the sets emerging from that step are therefore the claimed "preprocessed training data set" and "preprocessed test data set," Qiu applying the same normalization to the to-be-predicted sample before it enters the model, [0032], [0060].)
Qiu teaches "performing initialization training on the structural parameters to obtain an initialized model" (Qiu [0051]: "construction and initialization of a model BPNNs: the model is composed of an input layer, three hidden layers and an output layer, and the layers are fully connected"; Qiu [0018]: "the initialization of the hidden layer adopts the following method: weights are all initialized as random numbers between [-1,1] that obey normal distribution, and deviations are all initialized to 0"; Qiu Fig. 3: "Select different combinations of h1, h2 and h3 within the range L, construct a model BPNNs, and initialize the parameters of the hidden layers" – EN: setting the layer structure and then initializing the weights and deviations of those set structural parameters, together with fixing the learning rate, batch size, and number of learning cycles before iterative training ([0052]), is the claimed initialization training performed on the structural parameters to obtain an initialized model, consistent with the instant specification's description of initialization training at [0070]–[0071].) Qiu further evidences that "a network weight and a bias" is obtained through the subsequent training (Qiu [0052]: "the weights and the deviations are updated by using a small batch gradient descent method," the "deviations" being the biases initialized to 0 in Qiu [0018]).
Before the effective filing date of the claimed invention, it would have been obvious to a person having ordinary skill in the art to combine the deep neural network that predicts solar cell performance from structure data of Yi with the data preprocessing and the network initialization and weight-and-deviation training methodology of QIU. The motivation for doing so would be to obtain accurate predictions from imperfect collected data: preprocessing removes noise and scale differences from the data samples before they train the network, and initializing the weights and deviations and then updating them by gradient descent against a set error target is how a constructed network becomes a dependable trained prediction model. As QIU elaborates regarding the benefit of this preprocessing in paragraph 29, "First, to deal with the problems of noise, feature value missing and the like in the actual measurement data samples, the data denoising and data supplement processing is performed on the historical actual measurement data set; and then, the data normalization processing is performed on the historical actual measurement data set and the calculation simulation data set, respectively." And regarding the reliability of the resulting model, QIU states that its methodology is "thus improving the generalization ability of the performance prediction model, and effectively realizing the efficient and reliable prediction of the performance of the customized product in the design stage." (Para 19 of QIU).
Regarding claim 2,
Yi in view of Qiu teaches the method according to claim 1 as set forth above. Yi further teaches: "wherein the machine learning algorithm comprises at least one selected from the group consisting of a deep learning algorithm, a multilayer perceptron algorithm, a decision tree algorithm, a linear regression algorithm, a gradient boosting regression algorithm, and a k-nearest neighbor algorithm" (Yi, Abstract, p. 1: "In this paper, we propose a deep learning approach to predict and optimize cell performance of perovskite/crystalline-silicon (c-Si) tandem solar cells."; Yi, Sec. 2.2, p. 3: "a deep fully-connected neural network, named current prediction network (CPN), is designed and trained to predict the current properties of tandem solar cells with given layer thickness"). The recitation "at least one selected from the group consisting of" is in the alternative and is satisfied by a single member of the group; Yi's deep learning algorithm is such a member. Yi's deep fully connected feedforward network is additionally a multilayer perceptron, satisfying a second member of the group.
Regarding claim 3,
Yi in view of QIU teaches all the limitations of claim 2, Qiu further teaches:
"wherein the deep learning algorithm comprises at least one selected from the group consisting of a convolutional neural network algorithm, a self-encoding network algorithm, and a deep belief network algorithm. (Qiu [0013]: "a neural network model is trained by using the historical actual measurement data set and the calculation simulation data set to serve as the depth auto-encoder for the data samples. The depth auto-encoder is composed of an input layer, an encoder, a feature expression layer, a decoder and an output layer, and both the encoder and the decoder include three hidden layers"; Qiu [0009]: "encoding the historical actual measurement data set and the calculation simulation data set on the basis of a depth auto-encoder, mapping the data samples from an input space into a feature space, so as to express key features of the data samples"; Qiu [0060]: "inputting the data sample into the depth auto-encoder constructed in the step 3 for encoding, and finally inputting the encoded sample to be predicted into the prediction model BPNNsopt for prediction" -- EN: Qiu's depth auto-encoder corresponds to the claimed "self-encoding network algorithm," an auto-encoder being a neural network trained to encode its input into a feature space, and it is a deep learning algorithm as the claim requires because it is a deep neural network, a multi-hidden-layer architecture with three hidden layers in the encoder and three in the decoder trained on the data sets, so the "at least one selected from the group consisting of" recitation is satisfied.)
Before the effective filing date of the claimed invention, it would have been obvious to a person having ordinary skill in the art to combine the deep learning prediction of solar cell current properties from structure data of Yi with the depth auto-encoder of Qiu, so that the deep learning employed by the method comprises a self-encoding network algorithm that encodes the collected feature data before prediction. The motivation for doing so would be to express the key features of the data samples before they are used for prediction, so that the model receives the informative content of the samples rather than their raw form. As Qiu elaborates regarding the benefit of the depth auto-encoder in paragraph 19, "the method of the present application includes the steps of encoding the input feature of the data samples by using the depth auto-encoder, mapping the data samples from the input space into the feature space, so as to express key features of the data samples".
Regarding claim 9,
Yi in view of Qiu teaches the method according to claim 1 as set forth above. Yi further teaches
"wherein the optimizing the initialized model further comprises: evaluating a training result of the initialized model by using a mean square error (MSE)" (Yi, Sec. 3.1.1, p. 4: "Mean Square Error (MSE), Mean Absolute Percent Error (MAPE) and Mean Absolute Error (MAE) are considered as key performance metrics and defined in Table 2, where Oi and Pi denotes the ith observed value in FDTD data set and the ith prediction using CPN, respectively. MSE is chosen as the loss function for training"; Yi, Sec. 3.1.1, p. 4: "Fig. 5 illustrates how the MSE, MAPE and MAE of training and validation vary with the epochs."),
"wherein a formula of the mean square error is as follows: MSE = Σ(Predict_i - Actual_i)^2 / N" where Predict_i and Actual_i represent a predicted value and an actual value of an i-th sample, respectively, and N represents a data number in the preprocessed training data set. (Yi, Table 2, p. 4: "Mean Square Error (MSE) ... (1/n) Σ (Oi - Pi)^2"). – EN: Yi's formula is mathematically identical to the claimed formula: the squared difference (Oi - Pi)^2 equals (Pi - Oi)^2, Yi's prediction Pi corresponds to the claimed Predict_i, Yi's observed value Oi corresponds to the claimed Actual_i, and Yi's n corresponds to the claimed N, the number of samples over which the training-set MSE of Fig. 5 is computed, which in the combination set forth for claim 1 is the preprocessed training data set. Qiu further evidences the same evaluation (Qiu [0052]: "it is set that the mean square loss function is employed as the loss function during the training process").
Regarding claim 10,
Yi in view of Qiu teaches the method according to claim 1 as set forth above. Yi further teaches
The method according to claim 1, wherein the input feature parameters of the solar cell structures and the output feature parameters corresponding to the input feature parameters are screened and adjusted according to a type of each of the solar cell structures. (Yi, Sec. 2.1, p. 2: "To fully absorb the light transmitted to bottom sub-cell, the thickness of c-Si is often chosen to be as thick as 250 µm. Consequently, thicknesses of layers of top sub-cell determine the performance of the tandem solar cells"; Sec. 3.2.1, p. 5: the layer-thickness boundaries {Hg, Hi, Hm, Hp, He} are set to specified ranges "according to prior experiences" Yi, Sec. 2.4.1, p. 3: "The 2-terminal tandem solar cell is made up of top and bottom sub-cell in series circuit and the short-circuit current of tandem solar cells is bounded by the short-circuit current of any sub-cells"; Sec. 2.1, p. 2: "the thickness of each layer needs to be carefully designed and optimized so as high It and Ib and low Ir can be achieved" – EN: Yi screens and adjusts both feature sets to fit the perovskite/crystalline-silicon tandem cell type. On the input side, because this cell type has a thick, fully-absorbing c-Si bottom cell, the inputs are screened down to the five top sub-cell layer thicknesses that determine performance, and the range of each thickness is adjusted to values suited to this cell type. On the output side, because this cell type is a 2-terminal series-connected tandem whose current is bounded by the smaller sub-cell current, the outputs are selected as the top and bottom sub-cell currents It, Ib and the reflection loss Ir, the quantities that must be jointly high and low for this series configuration. Selecting and tuning the inputs and outputs to suit the perovskite/c-Si tandem type in this way corresponds to the input and output feature parameters being "screened and adjusted according to a type of each of the solar cell structures.)
Regarding claim 11,
Yi in view of Qiu teaches the method according to claim 1 as set forth above.
Yi further teaches:
"adjusting a design scheme of the to-be-predicted solar cell structure according to the predicted values of output feature parameters of the to-be-predicted solar cell structure" (Yi, Abstract, p. 1: "Heuristic algorithms are further adopted to inversely optimize the device structure, where the optimal set of layer thicknesses is obtained to maximize the achievable short-circuit current."; Yi, Sec. 2.4.1, p. 3: "Based on the predicted currents using CPN, let us now further search for the optimal layer thickness set {Hg, He, Hm, Hp, Hi} in real number space to maximize the smaller short-circuit current of top and bottom sub-cell"; Yi, Sec. 4, p. 6: "By searching in the whole integer space (in unit of nm), the optimal thickness set is found by SA, i.e., {Hg, He, Hm, Hp, Hi}={69, 15, 242, 10, 10} nm." – EN: Selecting the optimal set of layer thicknesses for the tandem structure on the basis of the currents predicted by the prediction model corresponds to adjusting the design scheme of the to-be-predicted solar cell structure according to the predicted values of its output feature parameters.)Claims 4 and 5 are rejected under 35 U.S.C. 103 as being unpatentable over Yi in view of QIU as applied to claim 1 above, and further in view of Derkacs et al. (US 10,263,134 B1), hereinafter "Derkacs".
Regarding claim 4,
Yi in view of Qiu teaches the method according to claim 1 as set forth above. Yi further teaches:
The method according to claim 1, wherein each of the solar cell structures is a multi-junction solar cell structure, and comprises at least one bottom cell (...) (Page 2, Section 2.1, "The perovskite/c-Si tandem solar cell structure is illustrated in Fig. 1. To fully absorb the light transmitted to bottom sub-cell, the thickness of c-Si is often chosen to be as thick as 250 µm." Page 1, Section 1, "Additionally, perovskite is an excellent top sub-cell in tandem solar cell structure owing to the flexible bandgap tuning of perovskite materials." - EN: this denotes Yi's tandem cell, which stacks two photovoltaic junctions, a perovskite top sub-cell above a c-Si bottom sub-cell, and is therefore a "multi-junction solar cell structure". The c-Si bottom sub-cell corresponds to the "at least one bottom cell", and the perovskite top sub-cell is a sub-cell disposed above it.)
Yi does not explicitly teach:
and a plurality of sub-cells, and the plurality of sub-cells are disposed above the bottom cell.
However, Derkacs teaches:
and a plurality of sub-cells, and the plurality of sub-cells are disposed above the bottom cell. (FIG. 3; Col. 6, lines 23-27, "As shown in an embodiment of a four junction upright lattice matched multijunction solar cell in the illustrated example of FIG. 3, the bottom subcell D includes a substrate 300 formed of p-type germanium ("Ge"), which also serves as a base layer." Col. 6, lines 35-40, "Heavily doped p-type aluminum gallium arsenide ("AlGaAs") and heavily doped n-type gallium arsenide ("GaAs") tunneling junction layers 303, 304 may be deposited over the nucleation layer to provide a low resistance pathway between the bottom and middle subcells." - EN: this denotes Derkacs's four junction solar cell of FIG. 3, in which subcells C, B and A are stacked, in that order, above the bottom subcell D. Subcells A, B and C correspond to the claimed "plurality of sub-cells", and the bottom subcell D corresponds to the claimed "bottom cell" above which they are disposed. In the combination, the solar cell structures whose data are collected and whose performance is predicted are multijunction structures of this kind.)
Before the effective filing date of the claimed invention, it would have been obvious to a person having ordinary skill in the art to combine the structure-to-performance prediction method for stacked solar cells of Yi with the multijunction solar cell having a bottom subcell and a plurality of subcells stacked above it of Derkacs, so that the structures whose data are collected and predicted carry a plurality of sub-cells above the bottom cell. The motivation for doing so would be to apply the prediction method to cell designs that convert more of the solar spectrum into electricity, since stacking additional subcells with different band gaps raises the attainable conversion efficiency. As Derkacs elaborates regarding the benefit of a plurality of subcells in Col. 1, lines 42-48, "The higher conversion efficiency of III-V compound semiconductor solar cells compared to silicon solar cells is in part based on the ability to achieve spectral splitting of the incident radiation through the use of a plurality of photovoltaic regions with different band gap energies, and accumulating the current from each of the regions."
Regarding claim 5,
Yi in view of Qiu further in view of Derkacs teaches the method according to claim 4 as set forth above.
Derkacs further teaches:
The method according to claim 4, wherein the bottom cell comprises a substrate, (Derkacs, FIG. 3; Col. 6, lines 23–27: "the bottom subcell D includes a substrate 300 formed of p-type germanium ('Ge'), which also serves as a base layer.")
an emissive layer (Derkacs, Col. 6, lines 28–30: "The bottom subcell D, further includes, for example, a highly doped n-type Ge emitter layer 301"; Col. 6, lines 32–35: "the emitter layer is formed in the substrate by diffusion of deposits into the Ge substrate, thereby forming the n-type Ge layer 301." – EN: emitter layer 301 is the emitter of subcell D's p-n junction; "emissive layer" is not separately defined in the specification, and under its broadest reasonable interpretation the emitter, the carrier-emitting layer of the photovoltaic junction, corresponds to the claimed "emissive layer.")
a window layer (Derkacs, Col. 7, lines 41–43: "a highly doped n-type indium aluminum phosphide ('AlInP2') window layer 309" [and similarly window layers 315, 321 over subcells B and A]; Col. 7, lines 47–51: the window layer "also helps reduce the recombination loss and improves passivation of the cell surface of the underlying junctions." – EN: Derkacs discloses a window layer over the emitter of each subcell (309, 315, 321); it would have been obvious to likewise deposit a window layer over the emitter of bottom subcell D to obtain the same recombination-reduction and passivation benefit Derkacs attributes to those window layers, so that the bottom cell comprises a substrate, an emissive layer, a window layer, and a tunnel junction arranged sequentially.)
and a tunnel junction (Derkacs, Col. 6, lines 35–40: "Heavily doped p-type aluminum gallium arsenide ('AlGaAs') and heavily doped n-type gallium arsenide ('GaAs') tunneling junction layers 303, 304 may be deposited over the nucleation layer to provide a low resistance pathway between the bottom and middle subcells." – EN: the tunneling junction layers 303, 304 deposited above subcell D provide the low-resistance interconnect between the bottom cell and the cells above it, corresponding to the claimed "tunnel junction.")
that are arranged sequentially in a stacking direction, (Derkacs, FIG. 3. – EN: FIG. 3 shows subcell D built up in the stacking direction in the order substrate 300, emitter layer 301, then tunneling junction 303/304 above, with the window layer added over the emitter, this corresponds to the substrate, emissive layer, window layer, and tunnel junction being "arranged sequentially in a stacking direction.")
and the plurality of sub-cells are disposed to stack on the tunnel junction of the bottom cell; and (Derkacs, Col. 6, lines 41–44: "Distributed Bragg reflector (DBR) layers 305 are then grown adjacent to and between the tunnel diode 303, 304 of the bottom subcell D and the third solar subcell C"; Col. 6, lines 46–52: "in other embodiments, the distributed Bragg reflector (DBR) layers may be located between tunnel diode layers 304/303 and buffer layer 302." – EN: subcells C, B and A are stacked above the tunnel junction 303/304 of bottom subcell D, corresponding to the plurality of sub-cells being "disposed to stack on the tunnel junction of the bottom cell.")
each of the plurality of sub-cells comprises a back surface field layer, (Derkacs, Col. 7, lines 36–38: "the subcell C includes a highly doped p-type aluminum gallium arsenide ('AlGaAs') back surface field ('BSF') layer 306"; Col. 7, lines 56–58 [subcell B, BSF 312]; Col. 8, lines 1–4 [subcell A, BSF 318]. – EN: each of subcells C, B and A includes a BSF layer at the bottom of the subcell (306, 312, 318), corresponding to the claimed "back surface field layer.")
a substrate region, (Derkacs, Col. 7, lines 38–39 [subcell C, "a p-type InGaAs base layer 307"]; Col. 7, lines 58–59 [subcell B, "a p-type AlGaAs base layer 313"]; Col. 8, lines 4–5 [subcell A, "a base layer 319"]. – EN: The claim uses "substrate" for the bottom cell and the distinct term "substrate region" for each sub-cell, denoting the bulk absorber body of the sub-cell situated between its back surface field layer and its emissive layer. Under BRI in light of the specification, which does not separately define "substrate region," Derkacs's base layers (307, 313, 319), which occupy exactly that position between the BSF layer and the emitter layer, correspond to the claimed "substrate region.")
an emissive layer (Derkacs, Col. 7, lines 39–41 [subcell C, "a highly doped n-type indium gallium arsenide ('InGaAs') emitter layer 308"]; Col. 7, lines 59–60 [subcell B, InGaP2 or AlGaAs emitter 314]; Col. 8, lines 4–8 [subcell A, emitter layer 320]. – EN: each sub-cell's emitter layer (308, 314, 320) is the emitter of the sub-cell's junction; under the broadest reasonable interpretation of the undefined term "emissive layer," the emitter, the carrier-emitting layer of the photovoltaic junction, corresponds to the claimed "emissive layer," consistent with the construction set forth for the bottom cell above.)
and a window layer (Derkacs, Col. 7, lines 41–43 [subcell C, "a highly doped n-type indium aluminum phosphide ('AlInP2') window layer 309"]; Col. 7, lines 60–61 [subcell B, AlGaAlP window 315]; Col. 8, lines 20–21 [subcell A, "A highly doped n-type InAlP2 window layer 321 is deposited over subcell A"]. – EN: each sub-cell carries a window layer over its emitter (309, 315, 321), corresponding to the claimed "window layer.")
that are arranged sequentially in the stacking direction; (Derkacs, FIG. 3. – EN: FIG. 3 shows each of subcells C, B and A built up in the stacking direction thus being "arranged sequentially in the stacking direction.")
and an upper sub-cell of the plurality of sub-cells further comprises a contact layer disposed on the corresponding window layer of the upper sub-cell, (Derkacs, FIG. 3; Col. 8, lines 20–25: "A highly doped n-type InAlP2 window layer 321 is deposited over subcell A. After the cap or contact layer 322 is deposited, the grid lines are formed via evaporation and lithographically patterned and deposited over the cap or contact layer 322." – EN: subcell A is the topmost subcell and corresponds to the claimed "upper sub-cell"; FIG. 3 shows contact layer 322 disposed directly on subcell A's window layer 321, corresponding to the claimed "contact layer disposed on the corresponding window layer of the upper sub-cell.")
and each of the remaining sub-cells comprises a tunnel junction disposed on the corresponding window layer thereof. (Derkacs, FIG. 3; Col. 7, lines 51–55: "Before depositing the layers of the subcell B, heavily doped n-type InGaP and p-type AlGaAs (or other suitable compositions) tunneling junction layers 310, 311 may be deposited over the subcell C"; Col. 7, lines 64–67: "Before depositing the layers of the top cell A, heavily doped n-type InGaP and p-type AlGaAs tunneling junction layers 316, 317 may be deposited over the subcell B." – EN: the remaining sub-cells C and B each carry a tunnel junction on their respective window layer, tunnel junction 310/311 over window 309 of subcell C, and tunnel junction 316/317 over window 315 of subcell B which corresponds to "each of the remaining sub-cells comprises a tunnel junction disposed on the corresponding window layer thereof.")
Before the effective filing date of the claimed invention, it would have been obvious to a person having ordinary skill in the art to combine the deep neural network that predicts solar cell performance from structure data of Yi in view of Qiu with the solar cell configuration of Derkacs in order to predict the photovoltaic performance of Derkacs's multijunction solar cell from its layer structure, and thereby identify the layer configurations that yield higher conversion efficiency without exhaustively fabricating or numerically simulating each candidate structure. The motivation for doing so would be that Derkacs itself teaches that the performance of a multijunction solar cell is governed by, and highly sensitive to, the specific choice and arrangement of its semiconductor layers, and that selecting among these layer configurations is a design objective of the reference. Derkacs states that "the characteristic of sunlight absorption in semiconductor material, also known as photovoltaic properties, is critical to determine the most efficient number and sequence of subcells, and the semiconductor material (with specific bandgap, thickness, and electrical properties) in each subcell, to achieve the optimum energy conversion" (Derkacs, Col. 1–2)
Claim 6 is rejected under 35 U.S.C. 103 as being unpatentable over Yi in view of QIU as applied to claim 1 above, and further in view of Derkacs, and further in view of Rueangnetr et al. ("Effect of quantum well width on the electron and hole states in different single quantum well structures", hereinafter "Rueangnetr".
Regarding claim 6, Yi in view of Qiu teaches the method according to claim 1 as set forth above.
Yi further teaches:
The method according to claim 1, wherein the input feature parameters of the solar cell structures comprise: a thickness of each layer of each of the solar cell structures, (...); and (Page 3, Section 2.3, "In this paper, we formulate a data set with of 12,500 sets of numerical simulation data from FDTD, which are paired data of the set of layer thickness (in unit of nm) and current properties of tandem solar cells." Page 2, Section 2.1, "Let Hg, He, Hm, Hp and Hi denote the layer thickness of glass, Indium Tin Oxide (ITO), [6,6]-phenyl-C61-butyric acid methyl ester (PCBM), MAPBI3 and Poly(3,4-ethylenedioxythiophene (PEDOT), respectively, from top to bottom of the top sub-cell." - EN: this denotes the thickness of each of the five layers of each cell in the data set being an input feature parameter, which corresponds to "a thickness of each layer of each of the solar cell structures".)
the output feature parameters corresponding to the input feature parameters comprise: short circuit current densities, (...) of the solar cell structures. (Page 1, Section 1, "According to the absorption spectrum, subsequently calculated maximum achievable short-circuit current density (MACD) and optimal parameters are found, such as the thickness of layers or specific structure parameters of nanostructures." Page 3, Section 2.3, "For each layer thickness set, we run the FDTD simulation of the tandem solar cell and obtain the corresponding MACD of top sub-cell It, MACD of bottom sub-cell Ib and reflection loss Ir." - EN: this denotes the output feature parameters including the MACDs It and Ib, which are short-circuit current densities (expressed in mA/cm2), corresponding to the claimed "short circuit current densities".)
Yi does not explicitly teach:
stacking modes between layers of the solar cell structures, a shape of each layer of each of the solar cell structures, composition materials of each layer of each of the solar cell structures, and component ratios of the composition materials
[output parameters comprise…] open circuit voltages, and fill factors [...of solar cell structures]
However, Derkacs teaches:
stacking modes between layers of the solar cell structures, (...) composition materials of each layer of each of the solar cell structures, and component ratios of the composition materials (Col. 2, lines 8-12, "Typical III-V compound semiconductor solar cells are fabricated on a semiconductor wafer in vertical, multijunction structures or stacked sequence of solar subcells, each subcell formed with appropriate semiconductor layers and including a p-n photoactive junction." Col. 8, lines 30-34, "In some embodiments, the multijunction solar cell is an upright multijunction solar cell. In some embodiments, the multijunction solar cell is an upright metamorphic multijunction solar cell. In some embodiments, multijunction solar cell is an inverted metamorphic solar cell." Col. 6, lines 15-17, "Thus, depending upon the desired band gap, the material constituents of ternary materials can be appropriately selected for growth." Col. 7, lines 43-45, "The InGaAs base layer 307 of the subcell C can include, for example, approximately 0.015 mole fraction indium. Other compositions may be used as well." Col. 7, lines 61-63, "The InGaP emitter layer 314 of the subcell B can include, for example, approximately 0.5 mole fraction indium." - EN: this denotes the design variables of Derkacs's cell: the order and manner in which the layers are stacked (upright, upright metamorphic, or inverted metamorphic; the stacked sequence of subcells), which corresponds to "stacking modes between layers"; the semiconductor material chosen for each layer (Ge, InGaAs, AlGaAs, InGaP, and so on), which corresponds to "composition materials of each layer"; and the mole fractions of the constituents within each material (for example 0.015 or 0.5 mole fraction indium), which correspond to "component ratios of the composition materials".)
[output parameters comprise…] open circuit voltages, and fill factors [...of solar cell structures] (Col. 4, lines 56-61, "Traditional multijunction designs that incorporate high bandgap n-type emitters that can have greater than 0.2 mole fraction aluminum (e.g., AlInGaP and AlGaAs) may result in degraded solar cell parameters including quantum efficiency (QE), efficiency (EFF), fill factor (FF), and/or open-circuit voltage (Voc)." Col. 4, lines 31-35, "Adding aluminum to these materials may make them more susceptible to oxygen incorporation during growth, which may result in reduced minority carrier lifetimes and poor material quality (quantum efficiency, voltage, higher Eg-Voc, low FF, etc.)." - EN: this denotes the open-circuit voltage (Voc) and fill factor (FF) as performance parameters of a solar cell that register the effect of structural and material design choices. In the combination they are output feature parameters predicted together with the short circuit current densities.)
Before the effective filing date of the claimed invention, it would have been obvious to a person having ordinary skill in the art to combine the thickness-based input features and current-density outputs of the prediction network of Yi with the stacking modes, composition materials, component ratios, open-circuit voltage and fill factor design-and-performance variables of Derkacs. The motivation for doing so would be to have the prediction model account for every design variable that sets the cell's energy conversion and to report every performance parameter that a design change can degrade, so that a candidate design can be judged on its full performance rather than on current alone. Yi further elaborates on this, stating "It is important to further extend the deep learning approach to include more diverse parameters for efficiency prediction and optimization of tandem solar cells." (Pages 6-7, Section 4 of Yi). As Derkacs elaborates regarding the benefit of accounting for these variables in Col. 2, lines 1-7, "As such, the characteristic of sunlight absorption in semiconductor material, also known as photovoltaic properties, is critical to determine the most efficient number and sequence of subcells, and the semiconductor material (with specific bandgap, thickness, and electrical properties) in each subcell, to achieve the optimum energy conversion."
Yi in view of Qiu in view of Derkacs does not explicitly teach:
a shape of each layer of each of the solar cell structures
However, Rueangnetr teaches:
a shape of each layer of each of the solar cell structures, (Abstract, "In this study the electron and hole states in Al0.33Ga0.67As/GaAs single quantum well structures including squared QW, step QW and tilted QW, have been theoretically studied by solving the Schrodinger equation in real space." Page 1, Section 1, "The shape of potential profile is also important. Not only a squared QW but also a parabolic or triangle QW is considered". Page 1, Section 1, "Here, we concentrate on the effect of potential shape on the electron and hole states. The squared, step-like and tilted QWs are studied for different parameters." - EN: this denotes the shape of the well layer of the structure, which may be squared, step-like or tilted, being a structural parameter that changes the electron and hole energy states of the device. The shape of the layer's potential profile corresponds to "a shape of each layer of each of the solar cell structures".)
Before the effective filing date of the claimed invention, it would have been obvious to a person having ordinary skill in the art to combine the structural input feature parameters of the prediction method of Yi in view of Qiu in view of Derkacs with the layer shape parameter of Rueangnetr. The motivation for doing so would be to include among the input features a structural parameter that is known from the physics of the device to change the electron and hole energy states, and with them the device behavior, so that the model can predict the performance of quantum-well-containing cell designs. As Rueangnetr elaborates regarding the benefit of studying these structural parameters on Page 1, Section 1, "The quantum well solar cell is one of the applications which is studied in order to design the optimized solar cell with high efficiency".
Claims 7 and 8 are rejected under 35 U.S.C. 103 as being unpatentable over Yi in view of QIU as applied to claim 1 above, and further in view of Quatieri et al. (US 20150112232 A1), hereinafter "Quatieri".
Regarding claim 7, Yi in view of Qiu teaches the method according to claim 1 as set forth above. Yi further teaches:
The method according to claim 1, wherein the preprocessing the training data set and the test data set comprises the following steps:
screening feature data, comprising: screening the input feature parameters of the solar cell structures, the output feature parameters corresponding to the input feature parameters, and the input feature parameters of the to-be-predicted solar cell structure according to known physical knowledge and a relationship between the feature data; (Page 2, Section 2.1, "To fully absorb the light transmitted to bottom sub-cell, the thickness of c-Si is often chosen to be as thick as 250 µm. Consequently, thicknesses of layers of top sub-cell determine the performance of the tandem solar cells." Page 3, Section 2.4.1, "The 2-terminal tandem solar cell is made up of top and bottom sub-cell in series circuit and the short-circuit current of tandem solar cells is bounded by the short-circuit current of any sub-cells." - EN: this denotes Yi selecting which feature data to keep using known physical knowledge and the relationships among the features: because physics dictates that a 250 µm c-Si layer absorbs the transmitted light fully, its thickness is screened out and only the five top sub-cell thicknesses are kept as input features, and because the series circuit ties the cell's output current to the smaller sub-cell current, the output features kept are the two sub-cell current densities and the reflection loss. The same screened feature set is used for the training structures and for the to-be-predicted structures, which corresponds to screening the recited three categories of feature data "according to known physical knowledge and a relationship between the feature data".)
Yi does not explicitly teach:
data processing, comprising: performing normalization processing on the screened feature data to obtain processed feature data; and
data recombination, comprising: transforming or combining the processed feature data in different dimensions to improve an expressive power or reduce a complexity of the processed feature data, and to covert the processed feature data with higher dimension into the processed feature data with lower dimension.
However, Quatieri teaches:
data processing, comprising: performing normalization processing on the screened feature data to obtain processed feature data; and (Para 86, "A critical step was to first normalize each of the features into standard units (zero mean unit variance), which allowed the variation of each feature to be considered relative to its baseline variation across the feature frames in all the sessions in the Training set." Para 95, "The principal component features were subsequently normalized to zero mean, unit standard deviation (again, with normalizing coefficients obtained from the Training set only) and applied to the Development set prior to the GMM-based multivariate regression, described in Section 4.1." - EN: this denotes normalizing every feature onto a standard scale before it is given to the machine learning algorithm, which corresponds to "performing normalization processing on the screened feature data", and the normalized features are the "processed feature data". Quatieri computes the normalization from the training data and applies it to the held-out data as well, so both the training data set and the test data set are normalized.)
data recombination, comprising: transforming or combining the processed feature data in different dimensions to improve an expressive power or reduce a complexity of the processed feature data, and to covert the processed feature data with higher dimension into the processed feature data with lower dimension. (Para 86, "The cross-correlation analysis produced, from each feature frame, a high dimensional feature vector of correlation matrix eigenvalues and covariance matrix power and entropy values." "The features within each domain were highly correlated, and so the final stage of feature construction was dimensionality reduction, using principal component analysis (PCA), into a smaller set of uncorrelated features." Para 95, "Next, PCA was applied independently to each feature domain, generating the following number of components per feature domain: 5 principal components for the formant domain, and 6 principal components for the delta-mel-cepstral domain." - EN: this denotes recombining the normalized feature data by principal component analysis. PCA combines the correlated features into a new, smaller set of uncorrelated components, that is, it transforms and combines the processed feature data in different dimensions, it reduces the complexity of the data, and it converts a feature vector with a higher dimension (hundreds of elements) into one with a lower dimension (5 or 6 components). This corresponds to the claimed "data recombination", including the recited conversion of the higher-dimension processed feature data into the lower dimension.)
Before the effective filing date of the claimed invention, it would have been obvious to a person having ordinary skill in the art to combine the screened solar cell feature data of Yi in view of QIU with the standard-units normalization and the PCA dimensionality reduction of feature data of Quatieri, applied to the feature data before they enter the machine learning algorithm. The motivation for doing so would be to place every feature on a common scale so that no feature dominates the model merely because of its raw units, and to shrink a correlated feature set into a smaller uncorrelated one so that the model trains on less redundant data. As Quatieri elaborates regarding the benefit of this feature processing in paragraph 86, "A critical step was to first normalize each of the features into standard units (zero mean unit variance), which allowed the variation of each feature to be considered relative to its baseline variation across the feature frames in all the sessions in the Training set." And regarding the reduction, "The features within each domain were highly correlated, and so the final stage of feature construction was dimensionality reduction, using principal component analysis (PCA), into a smaller set of uncorrelated features." (Para 86 of Quatieri).
Regarding claim 8, Yi in view of Qiu further in view of Quatieri teaches the method according to claim 7 as set forth above. Quatieri further teaches:
The method according to claim 7, wherein after performing normalization processing on the screened feature data, a mean value of the processed feature data is 0 and a standard deviation of the processed feature data is 1. (Para 86, "A critical step was to first normalize each of the features into standard units (zero mean unit variance), which allowed the variation of each feature to be considered relative to its baseline variation across the feature frames in all the sessions in the Training set." Para 95, "The principal component features were subsequently normalized to zero mean, unit standard deviation (again, with normalizing coefficients obtained from the Training set only) and applied to the Development set prior to the GMM-based multivariate regression, described in Section 4.1." - EN: this denotes the normalized feature data having "zero mean unit variance", meaning the mean value of the processed feature data is 0 and the variance is 1. A variance of 1 is the same as a standard deviation of 1, because the standard deviation is the square root of the variance. Para 95 states the same condition directly as "zero mean, unit standard deviation". This corresponds to the claimed mean value of 0 and standard deviation of 1 after the normalization processing.)
Before the effective filing date of the claimed invention, it would have been obvious to a person having ordinary skill in the art to combine the screened solar cell feature data of Yi in view of QIU with the zero-mean, unit-variance standardization of feature data taught by Quatieri. The motivation for doing so would be to place features that span different units and magnitudes onto a common scale, so that each feature's variation is weighted relative to its own baseline rather than its raw units, improving the reliability of the trained model. As Quatieri explains, normalizing the features into standard units of zero mean and unit variance "allowed the variation of each feature to be considered relative to its baseline variation across the feature frames in all the sessions in the Training set." (Quatieri, Para 86.)
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to NAYMUR RAHMAN ALI whose telephone number is (571)272-0007. The examiner can normally be reached Mon-Fri. 9:30-6:30 pm.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Alexey Shmatov can be reached at (571)270-3428. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/NAYMUR RAHMAN ALI/Examiner, Art Unit 2123
/ALEXEY SHMATOV/Supervisory Patent Examiner, Art Unit 2123
Prosecution Insights