公开(公告)号：US20220405561A1

公开(公告)日：2022-12-22

申请号：US17893450

申请日：2022-08-23

Applicant: SAMSUNG ELECTRONICS CO., LTD.

Inventor： Sangjeong LEE ,  Jihun OH ,  Meejeong PARK

IPC: G06N3/063 ,  G06F5/01

Abstract: An electronic device and a controlling method of an electronic device are provided. An electronic device recursively determines a plurality of layers of a neural network model. Weight data of first model information is recursively quantized to obtain a second neural network model. The recursive quantization begins with the weight data and determines an iteration count of a recursion. The recursion operates on error data, quantized weight data, scale data and quantized error data to obtain the iteration count. A first bit-width of the weight data is reduced to a second bit-width of the quantized weight data. The recursion may be performed on a per-layer basis. The weight data may be formulated in a floating-point format and the quantized weight data may be formulated in a fixed point format with an integer number of bits.

公开(公告)号：US20240412052A1

公开(公告)日：2024-12-12

申请号：US18811302

申请日：2024-08-21

Applicant: SAMSUNG ELECTRONICS CO., LTD.

Inventor： Meejeong PARK ,  Jihun OH ,  Youngrae CHO ,  Jonghun LEE

IPC: G06N3/0495 ,  G06N3/0464

Abstract: A data processing method for neural network quantization, includes: obtaining a quantized weight by quantizing a weight of a neural network; obtaining a quantization error that is a difference between the weight and the quantized weight; obtaining input data with respect to the neural network; obtaining a first convolution result by performing convolution on the quantized weight and the input data; obtaining a second convolution result by performing convolution on the quantization error and the input data; obtaining a scaled second convolution result by scaling the second convolution result based on bit shifting; and obtaining output data by using the first convolution result and the scaled second convolution result.

公开(公告)号：US20220392520A1

公开(公告)日：2022-12-08

申请号：US17569679

申请日：2022-01-06

Applicant: SAMSUNG ELECTRONICS CO., LTD.

Inventor： Daero KIM ,  Kyunghoi KOO ,  Sujeong KIM ,  Juyoung KIM ,  Sanghune PARK ,  Jiyeon PARK ,  Jihun OH ,  Kyoungwon LEE

IPC: G11C11/4096 ,  G11C11/4076

Abstract: A memory controller includes a first receiver configured to compare a read reference voltage with a piece of data received through a first data line and output a first piece of data; a first duty adjuster configured to adjust a duty of the first piece of data; a second receiver configured to compare the read reference voltage with a piece of data received through a second data line and output a second piece of data; a second duty adjuster configured to adjust a duty of the second piece of data; and a training circuit configured to perform a training operation on pieces of data received through a plurality of data lines, to obtain a target read reference voltage for each piece of data and correct a duty of each piece of data based on a level of the target read reference voltage for each piece of data.

公开(公告)号：US20250094375A1

公开(公告)日：2025-03-20

申请号：US18965046

申请日：2024-12-02

Applicant: Samsung Electronics Co., Ltd.

Inventor： Taekyung YEO ,  Sangyun HWANG ,  Sujeong KIM ,  Jihun OH ,  Joohee SHIN

IPC: G06F13/40 ,  G06F13/42

Abstract: A method of training a physical interface between a first device and a second device includes performing a first training of the physical interface by communicating with the second device by using a first candidate group of lanes from among a plurality of lanes; performing a second training of the physical interface by communicating with the second device by using a second candidate group of lanes from among the plurality of lanes, the second candidate group being different from the first candidate group; determining a lane group based on a result of the first training and a result of the second training; and setting the second device so that the determined lane group is used for the physical interface.

公开(公告)号：US20220121592A1

公开(公告)日：2022-04-21

申请号：US17475705

申请日：2021-09-15

Applicant: Samsung Electronics Co., Ltd.

Inventor： Taekyung YEO ,  Sangyun HWANG ,  Sujeong KIM ,  Jihun OH ,  Joohee SHIN

IPC: G06F13/40 ,  G06F13/42

Abstract: A method of training a physical interface between a first device and a second device includes performing a first training of the physical interface by communicating with the second device by using a first candidate group of lanes from among a plurality of lanes; performing a second training of the physical interface by communicating with the second device by using a second candidate group of lanes from among the plurality of lanes, the second candidate group being different from the first candidate group; determining a lane group based on a result of the first training and a result of the second training; and setting the second device so that the determined lane group is used for the physical interface.

公开(公告)号：US20210365779A1

公开(公告)日：2021-11-25

申请号：US17171290

申请日：2021-02-09

Applicant: Samsung Electronics Co., Ltd.

Inventor： Jihun OH ,  Sangjeong LEE ,  Meejeong PARK ,  Gaurav POONIWALA ,  Kiseok KWON

IPC: G06N3/08 ,  G06N3/063 ,  G06N3/04

Abstract: An electronic apparatus is provided. The electronic apparatus includes a memory configured to store an artificial intelligence (AI) model including a plurality of layers and a processor, and the AI model may include a plurality of weight values that are scaled based on shift scaling factors different by a plurality of channels included in each of the plurality of layers and quantized by the plurality of layers, and the processor may, based on receiving input data, in a neural network computation process for the input data, compute a channel-wise computation result with an inverse-scaled composite scale parameter based on a shift scaling factor corresponding to each channel.