Abstract: In a method of manufacturing a semiconductor device, an n-type source/drain epitaxial layer and a p-type source/drain epitaxial layer respectively formed, a dielectric layer is formed over the n-type source/drain epitaxial layer and the p-type source/drain epitaxial layer, a first opening is formed in the dielectric layer to expose a part of the n-type source/drain epitaxial layer and a second opening is formed in the dielectric layer to expose a part of the p-type source/drain epitaxial layer, and the n-type source/drain epitaxial layer and the p-type source/drain epitaxial layer respectively recessed. A recessing amount of the n-type source/drain epitaxial layer is different from a recessing amount of the p-type source/drain epitaxial layer.

Abstract: A method includes forming a dummy gate over a semiconductor fin; forming a source/drain epitaxial structure over the semiconductor fin and adjacent to the dummy gate; depositing an interlayer dielectric (ILD) layer to cover the source/drain epitaxial structure; replacing the dummy gate with a gate structure; forming a dielectric structure to cut the gate structure, wherein a portion of the dielectric structure is embedded in the ILD layer; recessing the portion of the dielectric structure embedded in the ILD layer; after recessing the portion of the dielectric structure, removing a portion of the ILD layer over the source/drain epitaxial structure; and forming a source/drain contact in the ILD layer and in contact with the portion of the dielectric structure.

Abstract: A personalized voice conversion system includes a cloud server and an intelligent device that communicates with the cloud server. The intelligent device upstreams an original voice signal to the cloud server. The cloud server converts the original voice signal into an intelligible voice signal based on an intelligible voice conversion model. The intelligent device downloads and plays the intelligible voice signal. Based on the original voice signal and the corresponding intelligible voice signal, the cloud server and the intelligent device train an off-line voice conversion model provided to the intelligent device. When the intelligent device stops communicating with the cloud server, the intelligent device converts a new original voice signal into a new intelligible voice signal based on the off-line voice conversion model and plays the new intelligible voice signal.

Abstract: A personalized voice conversion system includes a cloud server and an intelligent device that communicates with the cloud server. The intelligent device upstreams an original voice signal to the cloud server. The cloud server converts the original voice signal into an intelligible voice signal based on an intelligible voice conversion model. The intelligent device downloads and plays the intelligible voice signal. Based on the original voice signal and the corresponding intelligible voice signal, the cloud server and the intelligent device train an off-line voice conversion model provided to the intelligent device. When the intelligent device stops communicating with the cloud server, the intelligent device converts a new original voice signal into a new intelligible voice signal based on the off-line voice conversion model and plays the new intelligible voice signal.

Abstract: A variable-latency computing device includes a computing module, enabling units, a storage operation unit, and a detecting controller. The enabling units are divided into at least two groups. The storage operation unit includes word lines and bit lines. The enabling units enable the word line. The storage operation unit accumulates the data values corresponding to the bit lines and the enabled word line, thereby computing first accumulation values. The detecting controller controls the computing module to stop receiving the first accumulation values when the sum of the first accumulation values is higher than a threshold value and takes turns to turn off the at least two groups. The storage operation unit computes second accumulation values during different periods. The computing module receives and computes the second accumulation values corresponding to the at least two groups, so as to generate a computation value.

Abstract: A memory is suitable for performing a deep neural network operation. The memory includes: a processing unit and a weight unit. The processing unit includes a data input terminal and a data output terminal. The weight unit is configured to be coupled to the data input terminal of the processing unit. The weight unit includes an index memory and a mapping table. The index memory is configured to store multiple weight indexes. The mapping table is configured to respectively map the multiple weight indexes to multiple representative weight data.

Abstract: The present invention discloses a speculative lookahead processing device and method to enhance the statistical performance of datapaths. The method comprises steps: entering an input signal to at least two datapath units in a round-robin way; outputting the correct value at the Nth cycle, and acquiring the speculation value at the Mth cycle beforehand to start the succeeding computation, wherein M and N are natural numbers and M is smaller than N; comparing the speculation value with the correct value at the Nth cycle to determine whether the speculation is successful; if successful, excluding extra activities; if not successful, deleting the succeeding computation undertaken beforehand and restarting the succeeding computation with the correct value.

Abstract: A method and a device for retrieving a breathing signal is disclosed. Firstly, a triaxial acceleration signal is retrieved for a fixed period. Then, multivariate empirical mode decomposition (MEMD) is performed on the triaxial acceleration signal, so as to sequentially obtain a plurality of intrinsic mode functions (IMFs). Finally, an average inclination angle of the triaxial acceleration signal corresponding to each intrinsic mode function is sequentially calculated, and the intrinsic mode function corresponding to the average inclination angle within a predetermined angle range is added to a set. All the intrinsic mode functions are added up to obtain and output a breathing signal when an amount of the intrinsic mode functions in the set equals to a predetermined value being a natural number.

Abstract: The present invention discloses a speculative lookahead processing device and method to enhance the statistical performance of datapaths. The method comprises steps: entering an input signal to at least two datapath units in a round-robin way; outputting the correct value at the Nth cycle, and acquiring the speculation value at the Mth cycle beforehand to start the succeeding computation, wherein M and N are natural numbers and M is smaller than N; comparing the speculation value with the correct value at the Nth cycle to determine whether the speculation is successful; if successful, excluding extra activities; if not successful, deleting the succeeding computation undertaken beforehand and restarting the succeeding computation with the correct value.