Abstract: A method for enhancing kernel reparameterization of a non-linear machine learning model includes providing a predefined machine learning model, expanding a kernel of the predefined machine learning model with a non-linear network for convolution operation of the predefined machine learning model to generate the non-linear machine learning model, training the non-linear machine learning model, reparameterizing the non-linear network back to a kernel for convolution operation of the non-linear machine learning model to generate a reparameterized machine learning model, and deploying the reparameterized machine learning model to an edge device.

Abstract: The present application belongs to the technical field of biopharmaceutical production devices, and discloses a culture vessel carrier automatic slide mechanism and a shaker incubator, by means of which a shake flask/shake tube/well plate carrier assembly can automatically slide in and out relative to an incubator main body. The culture vessel carrier automatic slide mechanism comprises a slide protective housing and is provided with a culture vessel carrier bearing member. A first slide rail and a second slide rail are arranged below the culture vessel carrier bearing member. The second slide rail is arranged below the culture vessel carrier bearing member and located above the first slide rail, and the first slide rail is arranged inside the incubator main body on a fixed base that does not slide.

Abstract: An anechoic chamber and a construction method thereof are provided, the anechoic chamber includes a top surface, being a polygon; trapezoid surfaces, corresponding to edges of top surface, upper edge lengths of trapezoid surface being equal to edge lengths of top surface, trapezoid surfaces being connected to edges of top surface through the upper edges, the trapezoid surfaces being sequentially connected along a circumferential direction of top surface, and being at angle to the top surface; rectangular surfaces, corresponding to the trapezoid surfaces, upper edge lengths of rectangular surface being equal to lower edge lengths of trapezoid surface, rectangular surfaces being connected to the trapezoid surfaces through the upper edges, the rectangular surfaces being sequentially connected along a circumferential direction of the lower edges of trapezoid surfaces, and being perpendicular to the top surface; and an absorbing material, disposed on the top surface, the trapezoid surfaces and the rectangular surf

Abstract: A micro electro mechanical system (MEMS) device and a method for manufacturing the same are provided. The MEMS device includes a substrate, a polymer film on the substrate and having a lower surface facing toward the substrate, a cavity passing through the substrate, and coil structures on the substrate and in the polymer film. The polymer film includes a corrugation pattern on the lower surface of the polymer film. A portion of the polymer film is exposed in the cavity.

Abstract: A capacitor on a fin structure includes a fin structure. A dielectric layer covers the fin structure. A first electrode extension is embedded within the fin structure. A first electrode penetrates the dielectric layer and contacts the first electrode extension. A second electrode and a capacitor dielectric layer are disposed within the dielectric layer. The capacitor dielectric layer surrounds the second electrode, and the capacitor dielectric layer is between the second electrode and the first electrode extension.

Abstract: In aspects of the disclosure, a method, a system, and a computer-readable medium are provided. The method of building a kernel reparameterization for replacing a convolution-wise operation kernel in training of a neural network comprises selecting one or more blocks from tensor blocks and operations; and connecting the selected one or more blocks with the selected operations to build the kernel reparameterization. The kernel reparameterization has a dimension same as that of the convolution-wise operation kernel.

Abstract: A reparameterization method for initializing a machine learning model includes initializing a prefix layer of a first low dimensional layer in the machine learning model and a postfix layer of the first low dimensional layer, inverting the prefix layer to generate an inverse prefix layer of the first low dimensional layer, inverting the postfix layer to generate an inverse postfix layer of the first low dimensional layer, combining the inverse prefix layer, the first low dimensional layer and the inverse postfix layer to form a high dimensional layer, generating parallel operation layers from the high dimensional layer, and assigning initial weights to the parallel operation layers.

Abstract: A method for removing branches from trained deep learning models is provided. The method includes steps (i)-(v). In step (i), a trained model is obtained. The trained model has a branch structure involving one or more original convolutional layers and a shortcut connection. In step (ii), the shortcut connection is removed from the branch structure. In step (iii), a reparameterization model is built by linearly expanding each of the original convolutional layers into a reparameterization block in the reparameterization model. In step (iv), parameters of the reparameterization blocks are optimized by training the reparameterization model. In step (v), each of the optimized reparameterization blocks is transformed into a reparameterized convolutional layer to form a branchless structure that replaces the branch structure in the trained model.

Abstract: The present disclosure provides a privacy computing method based on homomorphic encryption, which includes steps as follows. The ciphertext data is received, where the ciphertext data has a floating-point homomorphic encryption data structure, and the floating-point homomorphic encryption data structure of the ciphertext data includes the ciphertext mantissa, exponent parameter and gain parameter. The gain parameter sets the precision of the floating point corresponding to the ciphertext mantissa. The exponent parameter is adapted to multiplication or division. The artificial intelligence model performs operations on the ciphertext data to return the ciphertext result.

Abstract: A device includes a semiconductor fin, and a gate stack on sidewalls and a top surface of the semiconductor fin. The gate stack includes a high-k dielectric layer, a work-function layer overlapping a bottom portion of the high-k dielectric layer, and a blocking layer overlapping a second bottom portion of the work-function layer. A low-resistance metal layer overlaps and contacts the work-function layer and the blocking layer. The low-resistance metal layer has a resistivity value lower than second resistivity values of both of the work-function layer and the blocking layer. A gate spacer contacts a sidewall of the gate stack.

Abstract: A semiconductor device structure is provided. The semiconductor device structure includes a substrate, a conductive feature on the substrate, and an electrical connection structure on the conductive feature. The electrical connection includes a first grain made of a first metal material, and a first inhibition layer made of a second metal layer that is different than the first metal material. The first inhibition layer extends vertically along a first side of a grain boundary of the first grain and laterally along a bottom of the grain boundary of the first grain.

Abstract: A manufacturing method of an embedded component package structure includes the following steps: providing a carrier and forming a semi-cured first dielectric layer on the carrier, the semi-cured first dielectric layer having a first surface; providing a component on the semi-cured first dielectric layer, and respectively providing heat energies from a top and a bottom of the component to cure the semi-cured first dielectric layer; forming a second dielectric layer on the first dielectric layer to cover the component; and forming a patterned circuit layer on the second dielectric layer, the patterned circuit layer being electrically connected to the component.

Abstract: The present disclosure provides a semiconductor structure with having a source/drain feature with a central cavity, and a source/drain contact feature formed in central cavity of the source/drain region, wherein the source/drain contact feature is nearly wrapped around by the source/drain region. The source/drain contact feature may extend to a lower most of a plurality semiconductor layers.

Abstract: A semiconductor device structure, along with methods of forming such, are described. The method includes forming a first dielectric feature between first and the second fin structures, wherein each first and second fin structure includes first semiconductor layers and second semiconductor layers alternatingly stacked and in contact with the first dielectric layer. The method also includes removing the second semiconductor layers so that the first semiconductor layers of the first and second fin structures extend laterally from a first side and a second side of the first dielectric feature, respectively, trimming the first dielectric feature so that the first dielectric feature has a reduced thickness on both first and the second sides, and forming a gate electrode layer to surround each of the first semiconductor layers of the first and second fin structures.

Abstract: The present disclosure provides a light-emitting package. The light-emitting package includes a main body, a cavity disposed in the cavity, a base plane in the cavity and a light-emitting element. The light-emitting element is disposed in the cavity and connected to the base plane. The light-emitting element includes a substrate and a semiconductor stack on the substrate. The substrate includes a side wall, and the side wall incudes a first cutting trace. The main body includes a step portion disposed in the cavity and it surrounds the light-emitting element. The step portion comprises a first height relative to base plane, and the first cutting trace comprises a second height relative to the base plane. The second height is greater than the first height.

Abstract: Disclosed are a power demand side speech interaction method and system. The method includes: obtaining original demand information, the original demand information including user's basic information, user demand information, and a user demand time; converting the original demand information into first information in text format; performing text statistical analysis based on an industry term on the first information in text format, to obtain second information; searching for corresponding user's actual information from a database according to the second information; outputting the user's actual information; searching for a corresponding forecasting model from the database, according to the second information and the user's basic information; calculating, according to a policy limit value of latest policy information in the database, a time for which the model corresponding to the user's basic information reaches the policy limit value; and transmitting an early warning message.

Abstract: Methods and devices that include a multigate device having a channel layer disposed between a source feature and a drain feature, a metal gate that surrounds the channel layer, and a first air gap spacer interposing the metal gate and the source feature and a second air gap spacer interposing the metal gate and the drain feature. A backside contact extends to the source feature. A power line metallization layer is connected to the backside contact.

Abstract: Provided is a method and a system for transmitting information. The method is applicable to a processing device, and includes: acquiring a target image of a display device; determining a target area in the target image; and sending display information to the display device, wherein the display information includes information of the target area; wherein the target area is a partial pixel area of the target image, the target area includes pixels with transparencies less than 1, and transparencies of the pixels outside the target area in the target image are all 1.

Abstract: Disclosed is a method of fabricating a contact in a semiconductor device. The method includes: receiving a semiconductor structure having an opening into which the contact is to be formed; forming a metal layer in the opening; forming a bottom anti-reflective coating (BARC) layer in the opening; performing implanting operations with a dopant on the BARC layer and the metal layer, the performing implanting operations including controlling an implant energy level and controlling an implant dosage level to form a crust layer with a desired minimum depth on top of the BARC layer; removing unwanted metal layer sections using wet etching operations, wherein the crust layer and BARC layer protect remaining metal layer sections under the BARC layer from metal loss during the wet etching operations; removing the crust layer and the BARC layer; and forming the contact in the opening over the remaining metal layer sections.

Abstract: A method of wireless communication, by a user equipment (UE), includes receiving multiple neural network training configurations for channel state feedback (CSF). Each configuration corresponds to a different neural network framework. The method also includes training each of a group of neural network decoder/encoder pairs in accordance with the received training configurations. A method of wireless communication, by a base station, includes transmitting multiple neural network training configurations to a user equipment (UE) for channel state feedback (CSF). Each configuration corresponds to a different neural network framework. The method also includes receiving a neural network decoder/encoder pair trained in accordance with the training configurations.