Tsinghua University

Kea-Tiong Tang

Hsin Chia Chen

Da-Jeng Yao

Chia-Min Yang

Syuan-Hao Sie

Yu-Po Lin

Meng-Fan Chang

Chih-Cheng Hsieh

Shih-Wen Chiu

Yi-Wen Liu

Jyuo-Min Shyu

Hsiang-Chiu Wu

Jye-Luen Lee

Ting-I Chou

Chung-Hung Shih

Li-Chun Wang

Wei-Chen Wei

A hardware/software co-compressed computing method for a static random access memory (SRAM) computing-in-memory-based (CIM-based) processing unit includes performing a data dividing step, a sparsity step, an address assigning step and a hardware decoding and calculating step. The data dividing step is performed to divide a plurality of kernels into a plurality of weight groups. The sparsity step includes performing a weight setting step. The weight setting step is performed to set each of the weight groups to one of a zero weight group and a non-zero weight group. The address assigning step is performed to assign a plurality of index codes to a plurality of the non-zero weight groups, respectively. The hardware decoding and calculating step is performed to execute an inner product to the non-zero weight groups and the input feature data group corresponding to the non-zero weight groups to generate the output feature data group.

Hardware/software co-compressed computing method and system for static random access memory computing-in-memory-based processing unit

A quantization parameter providing step of a quantization method is performed to provide a quantization parameter which includes a quantized input activation, a quantized weight and a splitting value. A parameter splitting step is performed to split the quantized weight and the quantized input activation into a plurality of grouped quantized weights and a plurality of grouped activations, respectively, according to the splitting value. A multiply-accumulate step is performed to execute a multiply-accumulate operation with one of the grouped quantized weights and one of the grouped activations, and then generate a convolution output. A convolution quantization step is performed to quantize the convolution output to a quantized convolution output according to a convolution target bit. A convolution merging step is performed to execute a partial-sum operation with the quantized convolution output according to the splitting value, and then generate an output activation.

Quantization method based on hardware of in-memory computing

A method for integrating a processing-in-sensor unit and an in-memory computing includes the following steps. A providing step is performed to transmit the first command signal and the initial data to the in-memory computing unit. A converting step is performed to drive the first command signal and the initial data to convert to a second command signal and a plurality of input data through a synchronizing module. A fetching step is performed to drive a frame difference module to receive the input data to fetch a plurality of difference data. A slicing step is performed to drive a bit-slicing module to receive the difference data and slice each of the difference data into a plurality of bit slices. A controlling step is performed to encode the difference address into a control signal, and the in-memory computing unit accesses each of the bit slices according to the control signal.

Method and system for integrating processing-in-sensor unit and in-memory computing unit

The present invention discloses a demodulating method for a demodulating circuit. The demodulating method comprises a step of utilizing a wave-shaping voltage step-down circuit to perform a wave-shaping process for a modulating input signal. A separate circuit of modulating input signal is filtering data zero of the modulating input signal and reserving data one of the modulating input signal. A triggering circuit of data recover is used to start the count of data span, until the data count is completed. A data recover circuit is used to recover the data.

Demodulating circuit and the method thereof

The present invention discloses a feedback type voltage regulator, including a voltage reference circuit for providing a reference voltage, distributed series feedback amplifiers electrically coupled to the voltage reference circuit and a power transistor which gate is electrically connected to the distributed series feedback amplifiers. The distributed series feedback amplifiers comprises three set of amplifiers serially connected with each other, wherein the relation of the gain of the first amplifier (A1), the second amplifier (A2), and the third amplifier (A3) is A1&#x3e;A2&#x3e;A3, wherein the relation of the bandwidth of the first amplifier (B1), the second amplifier (B2), and the third amplifier (B3) is B1&#x3c;B2&#x3c;B3.

Feedback type voltage regulator

A gas sensing system includes an air intake hole which comprises a plurality of micro-flow channels, a replaceable sensor for receiving an air from the air intake hole and detecting the received air, and a processing unit coupled to the replaceable sensor and doing determination for the component of the air according to a detection result of the replaceable sensor. Wherein, the replaceable sensor includes a plurality of sensing chips arranged in an array and each sensing chip is coated with a sensing thin film separately to detect a different gas.

Gas sensing system

A medical ventilator capable of early detecting and recognizing types of pneumonia, a gas recognition chip, and a method for recognizing gas thereof are disclosed. The gas recognition chip of the medical ventilator comprises a sensor array, a sensor interface circuit, a stochastic neural network chip, a memory and a microcontroller. The sensor array receives a plurality of multiple types of gases to produce odor signals corresponding to each type of gas. The sensor interface circuit analyzes the odor signals to produce gas pattern signals corresponding to each type of gas. The stochastic neural network chip amplifies the differences between the gas pattern signals and performs dimensional reduction on the gas pattern signals to aid the analysis. The memory stores training data. The microcontroller performs a mixed gas recognizing algorithm to early detect and recognize the type of the pneumonia according to the gas training data.

Growing community of inventors

Taipei, Taiwan

Kea-Tiong Tang