WO2020172952A1 - 一种存算一体芯片中数模转换电路与模数转换电路复用装置 - Google Patents

一种存算一体芯片中数模转换电路与模数转换电路复用装置 Download PDF

Info

Publication number
WO2020172952A1
WO2020172952A1 PCT/CN2019/081340 CN2019081340W WO2020172952A1 WO 2020172952 A1 WO2020172952 A1 WO 2020172952A1 CN 2019081340 W CN2019081340 W CN 2019081340W WO 2020172952 A1 WO2020172952 A1 WO 2020172952A1
Authority
WO
WIPO (PCT)
Prior art keywords
module
analog
multiplexer
conversion circuit
digital
Prior art date
Application number
PCT/CN2019/081340
Other languages
English (en)
French (fr)
Inventor
王绍迪
Original Assignee
北京知存科技有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 北京知存科技有限公司 filed Critical 北京知存科技有限公司
Priority to US17/434,082 priority Critical patent/US11990915B2/en
Publication of WO2020172952A1 publication Critical patent/WO2020172952A1/zh

Links

Images

Classifications

    • HELECTRICITY
    • H03ELECTRONIC CIRCUITRY
    • H03MCODING; DECODING; CODE CONVERSION IN GENERAL
    • H03M1/00Analogue/digital conversion; Digital/analogue conversion
    • H03M1/004Reconfigurable analogue/digital or digital/analogue converters
    • HELECTRICITY
    • H03ELECTRONIC CIRCUITRY
    • H03MCODING; DECODING; CODE CONVERSION IN GENERAL
    • H03M1/00Analogue/digital conversion; Digital/analogue conversion
    • H03M1/02Reversible analogue/digital converters
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06GANALOGUE COMPUTERS
    • G06G7/00Devices in which the computing operation is performed by varying electric or magnetic quantities
    • G06G7/12Arrangements for performing computing operations, e.g. operational amplifiers
    • G06G7/16Arrangements for performing computing operations, e.g. operational amplifiers for multiplication or division

Definitions

  • the invention relates to the field of microelectronic integrated circuits, in particular to a multiplexing device for a digital-to-analog conversion circuit and an analog-to-digital conversion circuit in a storage-computing integrated chip.
  • the storage-computing integrated chip architecture is currently considered to be one of the efficient hardware platforms for real-time intelligent processing of big data
  • the analog vector-matrix multiplication is the core circuit of the storage-computing integrated chip, especially the storage-computing integrated chip based on flash memory.
  • DAC digital-to-analog conversion circuit
  • ADC analog-to-digital conversion circuit
  • each input terminal of the analog vector-matrix multiplication circuit is required to be connected to a DAC, and each output terminal is connected to an ADC. As shown in Figure 2, this method greatly causes area waste and cost overhead.
  • the present invention provides a digital-to-analog conversion circuit and analog-to-digital conversion circuit multiplexing device in a storage-computing integrated chip, which shares DAC and ADC through time division multiplexing, thereby effectively reducing the number of components and circuit area , Reduce cost and expense, and facilitate integration.
  • a multiplexing device for a digital-to-analog conversion circuit and an analog-to-digital conversion circuit in a storage-computing integrated chip includes a first many-to-one multiplexer (M1-MUX) module, a digital-to-analog conversion circuit (DAC) module, a first For a multiplexer (1M-MUX) module; the input terminal of the DAC module is connected to the output terminal of the first M1-MUX module, and the output terminal of the DAC module is connected to the input terminal of the first 1M-MUX module;
  • M1-MUX first many-to-one multiplexer
  • DAC digital-to-analog conversion circuit
  • 1M-MUX first For a multiplexer
  • the first M1-MUX module includes a plurality of first M1-MUX units for selecting a signal from a plurality of input ports as output;
  • the first 1M-MUX module includes a plurality of first 1M-MUX units for outputting an input signal to different output ports;
  • the DAC module includes a plurality of DAC units for converting digital signals into analog signals; the input end of each DAC unit is connected to the output end of the corresponding first M1-MUX unit, and the output end of the DAC unit is connected to the corresponding first 1M -The input terminal of the MUX unit;
  • Each said DAC unit corresponds to a plurality of input ports, which are shared by the corresponding first M1-MUX unit and the corresponding first 1M-MUX unit in a time division multiplexing manner.
  • the multiplexing device also includes an analog vector-matrix multiplication circuit (AMAC) module and a switch transistor module; wherein the output terminal of the first 1M-MUX module is connected to the input terminal of the AMAC module through the switch transistor module;
  • AMAC analog vector-matrix multiplication circuit
  • the AMAC module is composed of programmable semiconductor devices, and the threshold voltage of each programmable semiconductor device can be dynamically programmed;
  • the switch transistor module includes a plurality of switch transistor units for turning on or disconnecting the signal connection between the first 1M-MUX unit and the input terminal of the AMAC module;
  • Each row of the AMAC module is connected to its corresponding switch transistor unit.
  • the multiplexing device further includes a second M1-MUX module, an analog-to-digital conversion circuit (ADC) module, and a second 1M-MUX module; wherein the input terminal of the ADC module is connected to the output terminal of the second M1-MUX, and the ADC module The output terminal of is connected to the input terminal of the second 1M-MUX module, and the input terminal of the second M1-MUX module is connected to the output terminal of the AMAC module.
  • ADC analog-to-digital conversion circuit
  • the second M1-MUX module includes a plurality of second M1-MUX units for selecting a signal from a plurality of input ports as output;
  • the ADC module includes multiple ADC units
  • the second 1M-MUX module includes a plurality of second 1M-MUX units for outputting an input signal to different output ports.
  • the controller also includes a controller, the controller and the first M1-MUX module, DAC module, first 1M-MUX module, AMAC module, switch transistor module, second M1-MUX module, ADC module, second 1M-MUX Module connection; used to control each time through the first and second M1-MUX unit to select a signal from multiple input signals as output; used to control each time through the first and second 1M-MUX unit to select an input
  • the signal is output to different output ports; used to control the opening and closing of the corresponding switch transistor unit at each moment.
  • the controller controls the switching transistor module and the first M1-MUX module and the first 1M-MUX module connected to the DAC module in a time division multiplexing manner, and selects the corresponding input signal to import to the AMAC module; the controller is time division multiplexed Mode control connects the second M1-MUX module and the second 1M-MUX module of the ADC module, and selects the corresponding output signal from the output terminal of the AMAC module for output.
  • the multiplexing device for the digital-to-analog conversion circuit and the analog-to-digital conversion circuit in the storage-calculation integrated chip provided by the present invention adopts specific The connection relationship and control of the DAC realizes the sharing of DAC and ADC units in a time-division multiplexing manner, thereby effectively reducing the number of components, reducing circuit area, reducing cost, and integrating.
  • Figure 1 is a block diagram of the analog vector-matrix multiplication circuit, the corresponding digital-to-analog conversion circuit and the analog-to-digital conversion circuit in a typical storage-computing integrated chip;
  • Figure 2 is an implementation diagram of the analog vector-matrix multiplication circuit, the corresponding digital-to-analog conversion circuit and the analog-to-digital conversion circuit in a typical storage-computing integrated chip;
  • FIG. 3 is a block diagram of an embodiment of a multiplexing device for a digital-to-analog conversion circuit and analog-to-digital conversion circuit in a storage-computing integrated chip according to the present invention
  • FIG. 4 is a circuit diagram of an embodiment of a digital-to-analog conversion circuit and analog-to-digital conversion circuit multiplexing device in a storage-computing integrated chip of the present invention
  • FIG. 5 is a flowchart of a method for multiplexing digital-to-analog conversion circuits in a storage-computing integrated chip according to the present invention
  • FIG. 6 is a flow chart of a multiplexing method of analog-to-digital conversion circuit in a storage-computing integrated chip according to the present invention
  • the analog vector-matrix multiplication circuit is the core circuit of the storage-computing integrated chip.
  • a typical AMAC because its processed signal is an analog signal, and usually its input signal and output signal are required to be a digital signal, therefore, it is usually necessary to convert the digital signal into an analog signal for input through a digital-to-analog conversion circuit (DAC) at the input.
  • DAC digital-to-analog conversion circuit
  • ADC analog-to-digital conversion circuit
  • each input end of the AMAC is required to be connected to a DAC, and each output end is connected to an ADC, as shown in Figure 2. This method greatly causes area waste and cost overhead.
  • an embodiment of the present invention provides a device for multiplexing a digital-to-analog conversion circuit and an analog-to-digital conversion circuit in a storage-computing integrated chip, as shown in FIG. 3, including a first M1-MUX module, DAC module, first 1M-MUX module, AMAC module, switch transistor module, second M1-MUX module, ADC module, second 1M-MUX module.
  • the AMAC includes: a programmable semiconductor device array
  • the principle of AMAC to implement analog vector-matrix multiplication is as follows: for a programmable semiconductor device array of M rows ⁇ N columns, the sources of all programmable semiconductor devices in each column are connected to the same analog voltage input terminal, and N columns Programmable semiconductor devices are correspondingly connected to N analog voltage input terminals, the gates of all programmable semiconductor devices in each row are connected to the same bias voltage input terminal, and M rows of programmable semiconductor devices are correspondingly connected to M bias voltage input terminals.
  • the drains of all programmable semiconductor devices in each column are connected to the same analog current output terminal, and N columns of programmable semiconductor devices are correspondingly connected to N analog current output terminals.
  • the threshold voltage of each programmable semiconductor device can be adjusted .
  • N is a positive integer greater than or equal to zero
  • M is a positive integer greater than or equal to zero.
  • M and N can be equal or unequal to form a topological structure of source coupling and drain summation.
  • each programmable semiconductor device can be regarded as a variable equivalent analog weight (denoted as W k,j , where 0 ⁇ k ⁇ M and 0 ⁇ j ⁇ N represent row number and column number respectively), which is equivalent to storing an analog data, while the programmable semiconductor device array stores an analog data array
  • the programmable semiconductor device array may also adopt a topological structure of gate coupling and source summation or a topological structure of gate coupling and drain summation, which is not limited in the embodiment of the present invention.
  • FIG. 3 and FIG. 4 it includes a first M1-MUX module, a DAC module, a first 1M-MUX module, AMAC module, switch transistor module, second M1-MUX module, ADC module, second 1M-MUX module.
  • the output terminal of the first M1-MUX module is connected with the input terminal of the DAC module
  • the output terminal of the DAC module is connected with the input terminal of the first 1M-MUX module
  • the output terminal of the first 1M-MUX module is connected to the input terminal of the first 1M-MUX module through the switching transistor module.
  • the input of the AMAC module is connected
  • the output of the AMAC module is connected to the input of the second M1-MUX module
  • the output of the second M1-MUX module is connected to the input of the ADC module
  • the output of the ADC module is connected to the second 1M -The input terminal of the MUX module is connected.
  • each DAC By sharing each DAC with multiple (such as T) input terminals of AMAC in a time-division multiplexed manner; similarly, each ADC is output by multiple (such as P) AMACs in a time-division multiplexed manner End sharing to reduce the number of ADCs and DACs, thereby reducing chip area.
  • T is usually a factor of 2 as M, for example, T takes 2, 4, 6, 8, etc.
  • P is usually a factor of N, for example, P takes 2, 4, 6, 8, etc.
  • the values of T and P are determined according to the actual chip area and delay.
  • the multiplexing method for DAC is as follows: when M digital signals need to be input to the AMAC module, the controller controls the first M1-MUX module and the first 1M-MUX module connected to the DAC module, and controls the corresponding switch transistors at the same time, time sharing Input digital signal.
  • the multiplexing method for the ADC is as follows (as shown in Figure 6):
  • the controller controls the second M1-MUX module and the second 1M-MUX connected to the ADC module
  • the module outputs the calculation result in time sharing.
  • it further includes: a programming circuit connected to the source, gate and/or substrate of each programmable semiconductor device for adjusting the threshold voltage of the programmable semiconductor device.
  • it further includes: a controller, which controls the multiplexing process of the DAC and the DAC.
  • the programmable semiconductor device may use floating gate transistors.
  • the analog vector-matrix multiplication circuit may further include: a conversion device, which is connected before the multiple analog voltage input terminals, and is used to convert the multiple analog current input signals into analog voltage input signals. , Output to the corresponding analog voltage input terminal.
  • a conversion device which is connected before the multiple analog voltage input terminals, and is used to convert the multiple analog current input signals into analog voltage input signals. , Output to the corresponding analog voltage input terminal.
  • each module of the present invention is not limited to the above-mentioned structure provided by the embodiment of the present invention, and may also be other structures known to those skilled in the art, which will not be described here. limited.
  • an embodiment of the present invention also provides a chip, which includes the above-mentioned analog vector-matrix multiplication circuit, a digital-to-analog conversion circuit, and an analog-to-digital conversion circuit multiplexing device.
  • an embodiment of the present invention also provides an electronic device, which may include the above-mentioned analog vector-matrix multiplication circuit, digital-to-analog conversion circuit, and analog-to-digital conversion circuit multiplexing device.
  • the electronic device may be, for example, Personal computer, laptop computer, cell phone, camera phone, smart phone, personal digital assistant, media player, navigation device, email device, game console, tablet computer, wearable device or any of these devices combination.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Power Engineering (AREA)
  • Software Systems (AREA)
  • Computer Hardware Design (AREA)
  • General Physics & Mathematics (AREA)
  • Analogue/Digital Conversion (AREA)

Abstract

一种存算一体芯片中数模转换电路与模数转换电路复用装置,包括数模转换电路(DAC)模块、模拟向量-矩阵乘法运算电路(AMAC)模块、模数转换电路(ADC)模块、第一多对一多路复用器(M1-MUX)模块、第二多对一多路复用器模块、第一一对多多路复用器(1M-MUX)模块、第二一对多多路复用器模块与开关晶体管模块。在AMAC输入端,每个DAC对应多个输入端口,通过第一M1-MUX模块与第一1M-MUX模块以时间复用的方式进行共享;在AMAC输出端,每个ADC对应多个输出端口,通过第二M1-MUX模块与第二1M-MUX模块以时间复用的方式进行共享;减小DAC与ADC的数量,减小芯片面积。

Description

一种存算一体芯片中数模转换电路与模数转换电路复用装置
本申请要求于2019年02月26日提交中国专利局、申请号为201910143140.7申请名称为“一种存算一体芯片中数模转换电路与模数转换电路复用装置”的中国专利申请的优先权,其全部内容通过引用结合在本申请中。
技术领域
本发明涉及微电子集成电路领域,尤其涉及一种存算一体芯片中数模转换电路与模数转换电路复用装置。
背景技术
存算一体芯片架构目前被认为是解决大数据实时智能处理的高效硬件平台之一,而模拟向量-矩阵乘法是存算一体芯片的核心电路,尤其是基于闪存的存算一体芯片。对于典型的模拟向量-矩阵乘法电路,由于其处理的信号是模拟信号,而通常其输入信号与输出信号要求是数字信号,因此,通常需要在输入端通过数模转换电路(DAC)把数字信号转换成模拟信号,而在输出端通过模数转换电路(ADC)把处理完之后的模拟信号转换成数字信号,如图1所示。
但是,高精度DAC和ADC的面积通常都很大,在典型的存算一体芯片中,要求模拟向量-矩阵乘法电路的每个输入端都连接一个DAC,而每个输出端都连接一个ADC,如图2所示,这种方式极大地造成面积浪费与成本开销。
发明内容
有鉴于此,本发明提供一种存算一体芯片中数模转换电路与模数转 换电路复用装置,通过时分复用的方式来共享DAC和ADC,从而有效减少元器件数量,减小电路面积,降低成本开销,利于集成化。
为了实现上述目的,本发明采用如下技术方案:
一种存算一体芯片中数模转换电路与模数转换电路复用装置,包括第一多对一多路复用器(M1-MUX)模块、数模转换电路(DAC)模块、第一一对多多路复用器(1M-MUX)模块;其中DAC模块的输入端连接第一M1-MUX模块的输出端,DAC模块的输出端连接第一1M-MUX模块的输入端;
所述的第一M1-MUX模块包括多个第一M1-MUX单元,用于从多个输入端口中选择一个信号作为输出;
所述的第一1M-MUX模块包括多个第一1M-MUX单元,用于把一个输入信号输出到不同的输出端口;
所述的DAC模块包括多个DAC单元,用于把数字信号转换为模拟信号;各个DAC单元的输入端连接相应的第一M1-MUX单元输出端,DAC单元的输出端连接相应的第一1M-MUX单元的输入端;
所述的每个DAC单元对应多个输入端口,通过相应的第一M1-MUX单元与相应的第一1M-MUX单元以时分复用的方式进行共享。
进一步地,复用装置还包括模拟向量-矩阵乘法运算电路(AMAC)模块与开关晶体管模块;其中,第一1M-MUX模块的输出端通过开关晶体管模块连接AMAC模块的输入端;
所述的AMAC模块由可编程半导体器件组成,每个可编程半导体器件的阈值电压可以动态编程;
所述的开关晶体管模块包括多个开关晶体管单元,用于导通或者断开第一1M-MUX单元与AMAC模块输入端的信号连接;
所述的AMAC模块的每一行分别连接其对应的开关晶体管单元。
进一步地,复用装置还包括第二M1-MUX模块、模数转换电路(ADC) 模块、第二1M-MUX模块;其中,ADC模块的输入端连接第二M1-MUX的输出端,ADC模块的输出端连接第二1M-MUX模块的输入端,第二M1-MUX模块的输入端连接AMAC模块的输出端。
所述的第二M1-MUX模块包括多个第二M1-MUX单元,用于从多个输入端口中选择一个信号作为输出;
所述的ADC模块包括多个ADC单元;
所述的第二1M-MUX模块包括多个第二1M-MUX单元,用于把一个输入信号输出到不同的输出端口。
进一步地,还包括控制器,控制器与第一M1-MUX模块、DAC模块、第一1M-MUX模块、AMAC模块、开关晶体管模块、第二M1-MUX模块、ADC模块、第二1M-MUX模块连接;用于控制每个时刻通过第一和第二M1-MUX单元从多个输入信号中选择一个信号作为输出;用于控制每个时刻通过第一和第二1M-MUX单元选择一个输入信号输出到不同的输出端口;用于控制每个时刻相应开关晶体管单元的开闭。
所述控制器按时分复用的方式控制开关晶体管模块以及连接DAC模块的第一M1-MUX模块与第一1M-MUX模块,选择相应的输入信号导入到AMAC模块;控制器按时分复用的方式控制连接ADC模块的第二M1-MUX模块与第二1M-MUX模块,选择AMAC模块输出端相应的输出信号进行输出。
本发明提供的存算一体芯片中数模转换电路与模数转换电路复用装置,通过设置第一和第二M1-MUX模块、开关晶体管模块以及第一和第二1M-MUX模块,采用特定的连接关系和控制,实现以时分复用的方式共享DAC和ADC单元,从而有效减少元器件数量,减小电路面积,降低成本开销,利于集成化。
为让本发明的上述和其他目的、特征和优点能更明显易懂,下文特举较佳实施例,并配合所附图式,作详细说明如下。
附图说明
为了更清楚地说明本发明实施例或现有技术中的技术方案,下面将对实施例或现有技术描述中所需要使用的附图作简单地介绍,显而易见地,下面描述中的附图仅仅是本发明的一些实施例,对于本领域普通技术人员来讲,在不付出创造性劳动的前提下,还可以根据这些附图获得其他的附图。
图1为典型存算一体芯片中模拟向量-矩阵乘法运算电路、相应的数模转换电路与模数转换电路结构框图;
图2为典型存算一体芯片中模拟向量-矩阵乘法运算电路、相应的数模转换电路与模数转换电路实施图;
图3为本发明一种存算一体芯片中数模转换电路与模数转换电路复用装置实施例框图;
图4为本发明一种存算一体芯片中数模转换电路与模数转换电路复用装置实施例电路图;
图5为本发明一种存算一体芯片中数模转换电路复用方法流程图;
图6为本发明一种存算一体芯片中模数转换电路复用方法流程图;
具体实施方式
下面将结合本发明实施例中的附图,对本发明实施例中的技术方案进行清楚、完整地描述,显然,所描述的实施例仅仅是本发明一部分实施例,而不是全部的实施例。基于本发明中的实施例,本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例,都属于本发明保护的范围。
模拟向量-矩阵乘法电路(AMAC)是存算一体芯片的核心电路。典型AMAC,由于其处理的信号是模拟信号,而通常其输入信号与输出信号要 求是数字信号,因此,通常需要在输入端通过数模转换电路(DAC)把数字信号转换成模拟信号进行输入,而在输出端通过模数转换电路(ADC)把处理完之后的模拟信号转换成数字信号进行输出,如图1所示。这样一来,要求AMAC的每个输入端都连接一个DAC,而每个输出端都连接一个ADC,如图2所示,这种方式极大地造成面积浪费与成本开销。
为解决现有技术中的上述技术问题,本发明实施例提供一种存算一体芯片中数模转换电路与模数转换电路复用装置,如图3所示,包括第一M1-MUX模块、DAC模块、第一1M-MUX模块、AMAC模块、开关晶体管模块、第二M1-MUX模块、ADC模块、第二1M-MUX模块。
该AMAC包括:可编程半导体器件阵列;
其中,AMAC实现模拟向量-矩阵乘法运算的原理如下:针对一个M行×N列的可编程半导体器件阵列,每一列的所有可编程半导体器件的源极均连接至同一模拟电压输入端,N列可编程半导体器件对应连接N个模拟电压输入端,每一行的所有可编程半导体器件的栅极均连接至同一偏置电压输入端,M行可编程半导体器件对应连接M个偏置电压输入端,每一列的所有可编程半导体器件的漏极均连接至同一个模拟电流输出端,N列可编程半导体器件对应连接N个模拟电流输出端,其中,每个可编程半导体器件的阈值电压均可调节。N为大于等于零的正整数,M为大于等于零的正整数,M和N可以相等,也可以不等,以此形成源极耦合、漏极求和的拓扑结构。
其中,通过预先按照一定规则动态调节各可编程半导体器件的阈值电压V TH,可将各可编程半导体器件看作一个可变的等效模拟权重(记为W k,j,其中0<k<M和0<j<N分别代表行号和列号),相当于存储一个模拟数据,而可编程半导体器件阵列则存储一个模拟数据阵列
Figure PCTCN2019081340-appb-000001
电路工作时,将一行模拟电压信号V 1~V N分别施加至N列可编程半导体器件,其中第K列所有可编程半导体器件的源极均得到一模拟电压信号V k,栅极输入一偏置电压V b,漏极分别输出电流信号I k,1~I k,N,其中,根据可编程半导体器件特性,I=V×W每个可编程半导体器件的漏极输出电流等于源极电压乘以该可编程半导体器件的权重,即I k,1=V kW k,1,I k,N=V kW k,N,因为每一列的所有可编程半导体器件的漏极均连接至同一个模拟电流输出端,根据基尔霍夫定律,所以在该模拟电流输出端的电流I j为该列所有可编程半导体器件的漏极电流之和,即为
Figure PCTCN2019081340-appb-000002
Figure PCTCN2019081340-appb-000003
多个模拟电流输出端输出多个电流和
Figure PCTCN2019081340-appb-000004
实现矩阵乘法运算功能。
当然,该可编程半导体器件阵列也可以采用栅极耦合、源极求和的拓扑结构或栅极耦合、漏极求和的拓扑结构,本发明实施例对此不作限制。
下面,对本发明实施例数模转换电路与模数转换电路复用装置的原理进行说明:如图3和图4所示,包括第一M1-MUX模块、DAC模块、第一1M-MUX模块、AMAC模块、开关晶体管模块、第二M1-MUX模块、ADC模块、第二1M-MUX模块。其中,第一M1-MUX模块的输出端与DAC模块的输入端连接,DAC模块的输出端与第一1M-MUX模块的输入端连接,第一1M-MUX模块的输出端通过开关晶体管模块与AMAC模块的输入端连接,AMAC模块的输出端与第二M1-MUX模块的输入端连接,第二M1-MUX模块的输出端与ADC模块的输入端连接,ADC模块的输出端与第二1M-MUX模 块的输入端连接。
通过把每个DAC,以时分复用的方式被AMAC多个(例如T个)输入端进行共享;同理,把每个ADC,以时分复用的方式被AMAC多个(例如P个)输出端进行共享,以此来减小ADC与DAC的数量,从而减小芯片面积。这里T通常为2为M的因子,例如,T取2,4,6,8,…等;这里P通常为N的因子,例如,P取2,4,6,8,…等。T和P的取值根据实际芯片的面积与延迟等决定。
对于DAC的复用方法如下:当M个数字信号需要输入到AMAC模块时,控制器控制连接DAC模块的第一M1-MUX模块与第一1M-MUX模块,同时控制相应的开关晶体管,分时输入数字信号。具体如下(如图5所示):在第一个时刻,选择第1,T+1,2T+1,…,(M/T-1)T+1个数字信号,同时导通AMAC模块相对应行的开关晶体管单元(其他行的开关晶体管单元关断),数字信号通过DAC模块转换成模拟信号,输入AMAC模块的第1,T+1,2T+1,…,(M/T-1)T+1行;然后关断第1,T+1,2T+1,…,(M/T-1)T+1行的开关晶体管单元,钳制住当前行的输入信号;在第二个时刻,选择第2,T+2,2T+2,…,(M/T-1)T+2个数字信号,同时导通AMAC模块相对应行的开关晶体管单元(其他行的开关晶体管单元关断),数字信号通过DAC模块转换成模拟信号,输入AMAC模块的第2,T+2,2T+2,…,(M/T-1)T+2行;然后关断第2,T+2,2T+2,…,(M-1)/T+2的开关晶体管单元,钳制住当前行的输入信号;依此类推,直到所有数字信号都输入到AMAC模块;最后执行模拟向量-矩阵乘法运算。
对于ADC的复用方法如下(如图6所示):当AMAC模块需要输出N个模拟向量-矩阵乘法运算结果时,控制器控制连接ADC模块的第二M1-MUX模块与第二1M-MUX模块分时输出运算结果。在第一个时刻,选择AMAC模块第1,P+1,2P+1,…,(N/P-1)/P+1列的输出端口,相应的模 拟信号通过ADC模块转换成数字信号进行输出;在第二个时刻,选择AMAC模块第2,P+2,2P+2,…,(N/P-1)/P+2列的输出端口,相应的模拟信号通过ADC模块转换成数字信号进行输出;依此类推,直到AMAC模块所有列的模拟信号都通过ADC模块转换成数字信号输出。
若是采用现有方案,我们需要在每一行连接一个DAC单元,每一列连接一个ADC单元,则需要M个DAC单元与N个ADC单元,通常M和N都比较大,例如1024。而采用本发明提供的方案,我们只需要M/T个DAC单元,N/P个ADC单元,(M/T)个第一M1-MUX单元,(N/P)个第二M1-MUX单元,(M/T)个第一1M-MUX单元,(N/P)个第二1M-MUX单元与M个开关晶体管单元。值得注意的是,每个(第一和第二)M1-MUX单元与(第一和第二)1M-MUX单元的面积远远小于ADC单元与DAC单元的面积。本领域技术人员可以理解的是,M和N越大,同时T和P越大,本发明的优势越明显。
在一个可选的实施例中,还包括:编程电路,连接每一个可编程半导体器件的源极、栅极和/或衬底,用于调控可编程半导体器件的阈值电压。
在一个可选的实施例中,还包括:控制器,控制上述DAC与DAC的复用过程。
在上述实施例中,该可编程半导体器件可以采用浮栅晶体管。
在一个可选的实施例中,该模拟向量-矩阵乘法运算电路还可以包括:转换装置,连接在多个模拟电压输入端之前,用于将多个模拟电流输入信号分别转换为模拟电压输入信号,输至对应的模拟电压输入端。
以上仅是举例说明本发明各模块的具体结构,在具体实施时,上述各模块的具体结构不限于本发明实施例提供的上述结构,还可以是本领域技术人员可知的其他结构,在此不作限定。
另一方面,本发明实施例还提供一种芯片,该芯片包括上述模拟向量-矩阵乘法运算电路、数模转换电路与模数转换电路复用装置。
另外,本发明实施例还提供一种电子设备,该电子设备可以包括上述的模拟向量-矩阵乘法运算电路、数模转换电路与模数转换电路复用装置,更具体的,电子设备例如可以为个人计算机、膝上型计算机、蜂窝电话、相机电话、智能电话、个人数字助理、媒体播放器、导航设备、电子邮件设备、游戏控制台、平板计算机、可穿戴设备或者这些设备中的任何设备的组合。
还需要说明的是,术语“包括”、“包含”或者其任何其他变体意在涵盖非排他性的包含,从而使得包括一系列要素的过程、方法、商品或者设备不仅包括那些要素,而且还包括没有明确列出的其他要素,或者是还包括为这种过程、方法、商品或者设备所固有的要素。在没有更多限制的情况下,由语句“包括一个……”限定的要素,并不排除在包括所述要素的过程、方法、商品或者设备中还存在另外的相同要素。
本发明中应用了具体实施例对本发明的原理及实施方式进行了阐述,以上实施例的说明只是用于帮助理解本发明的方法及其核心思想;同时,对于本领域的一般技术人员,依据本发明的思想,在具体实施方式及应用范围上均会有改变之处,综上所述,本说明书内容不应理解为对本发明的限制。

Claims (7)

  1. 一种存算一体芯片中数模转换电路与模数转换电路复用装置,其特征在于,包括第一多对一多路复用器模块、数模转换电路模块、第一一对多多路复用器模块;其中数模转换电路模块的输入端连接第一多对一多路复用器模块的输出端,数模转换电路模块的输出端连接第一一对多多路复用器模块的输入端;
    所述的第一多对一多路复用器模块包括多个多对一多路复用器单元;
    所述的数模转换电路模块包括多个数模转换电路单元;
    所述的第一一对多多路复用器模块包括多个第一一对多多路复用器单元。
  2. 根据权利要求1所述的复用装置,其特征在于,每个数模转换电路单元对应多个输入端口,通过相应的第一多对一多路复用器单元与相应的第一一对多多路复用器单元以时分复用的方式进行共享。
  3. 根据权利要求1所述的复用装置,其特征在于,还包括模拟向量-矩阵乘法运算电路模块与开关晶体管模块;其中,第一一对多多路复用器模块的输出端通过开关晶体管模块连接模拟向量-矩阵乘法运算电路模块的输入端;
    所述的模拟向量-矩阵乘法运算电路模块由可编程半导体器件组成,每个可编程半导体器件的阈值电压可以动态编程;
    所述的开关晶体管模块包括多个开关晶体管单元;
    所述的模拟向量-矩阵乘法运算电路模块的每一行分别连接其对应的开关晶体管单元。
  4. 根据权利要求1所述的复用装置,其特征在于,还包括第二多对一多路复用器模块、模数转换电路模块、第二一对多多路复用器模块; 其中,模数转换电路模块的输入端连接第二多对一多路复用器模块的输出端,模数转换电路模块的输出端连接第二一对多多路复用器模块的输入端,第二多对一多路复用器模块的输入端连接模拟向量-矩阵乘法运算电路模块的输出端;
    所述的第二多对一多路复用器模块包括多个多对一多路复用器单元;
    所述的模数转换电路模块包括多个模数转换电路单元;
    所述的第二一对多多路复用器模块包括多个第二一对多多路复用器单元。
  5. 根据权利要求4所述的复用装置,其特征在于,每个模数转换电路单元对应多个输出端口,通过相应的第二多对一多路复用器单元与相应的第二一对多多路复用器单元以时分复用的方式进行共享。
  6. 根据权利要求1所述的复用装置,其特征在于,还包括控制器,控制器与第一多对一多路复用器模块、数模转换电路模块、第一一对多多路复用器模块、模拟向量-矩阵乘法运算电路模块、开关晶体管模块、第二多对一多路复用器模块、模数转换电路模块、第二一对多多路复用器模块连接;用于控制每个时刻通过第一和第二多对一多路复用器单元从多个输入信号中选择一个信号作为输出;用于控制每个时刻通过第一和第二一对多多路复用器单元选择一个输入信号输出到不同的输出端口;用于控制每个时刻相应开关晶体管单元的开闭。
  7. 根据权利要求6所述的复用装置,其特征在于,控制器按时分复用的方式控制开关晶体管模块以及连接数模转换电路模块的第一多对一多路复用器模块与第一一对多多路复用器模块,选择相应的输入信号导入到模拟向量-矩阵乘法运算电路模块;控制器按时分复用的方式控制连接模数转换电路模块的第二多对一多路复用器模块与第二一对多多路复用器模块,选择模拟向量-矩阵乘法运算电路模块输出端相应的输出信号 进行输出。
PCT/CN2019/081340 2019-02-26 2019-04-03 一种存算一体芯片中数模转换电路与模数转换电路复用装置 WO2020172952A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US17/434,082 US11990915B2 (en) 2019-02-26 2019-04-03 Multiplexing device for digital-to-analog conversion circuit and analog-to-digital conversion circuit in storage and calculation integrated chip

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201910143140.7A CN111614353A (zh) 2019-02-26 2019-02-26 一种存算一体芯片中数模转换电路与模数转换电路复用装置
CN201910143140.7 2019-02-26

Publications (1)

Publication Number Publication Date
WO2020172952A1 true WO2020172952A1 (zh) 2020-09-03

Family

ID=72201294

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/081340 WO2020172952A1 (zh) 2019-02-26 2019-04-03 一种存算一体芯片中数模转换电路与模数转换电路复用装置

Country Status (3)

Country Link
US (1) US11990915B2 (zh)
CN (1) CN111614353A (zh)
WO (1) WO2020172952A1 (zh)

Families Citing this family (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111614353A (zh) * 2019-02-26 2020-09-01 北京知存科技有限公司 一种存算一体芯片中数模转换电路与模数转换电路复用装置
CN113109856A (zh) * 2021-04-06 2021-07-13 明峰医疗系统股份有限公司 基于时分复用技术的ct探测器动态采样方法、系统及计算机可读存储介质
CN113364462B (zh) * 2021-04-27 2022-09-02 北京航空航天大学 模拟存算一体多比特精度实现结构
CN113517009A (zh) * 2021-06-10 2021-10-19 上海新氦类脑智能科技有限公司 存算一体式智能芯片、控制方法及控制器
CN115906735B (zh) * 2023-01-06 2023-05-05 上海后摩智能科技有限公司 基于模拟信号的多比特数存算一体电路、芯片及计算装置

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20040113177A1 (en) * 2002-12-09 2004-06-17 John Wynne Multi-channel integrated circuit comprising a plurality of DACs, and a method for monitoring the output of the DACs
CN104300984A (zh) * 2014-10-21 2015-01-21 上海玮舟微电子科技有限公司 一种模数转换器和模数转换方法
CN108009640A (zh) * 2017-12-25 2018-05-08 清华大学 基于忆阻器的神经网络的训练装置及其训练方法
CN108777155A (zh) * 2018-08-02 2018-11-09 北京知存科技有限公司 闪存芯片
CN208507187U (zh) * 2018-08-02 2019-02-15 北京知存科技有限公司 闪存芯片
CN209390036U (zh) * 2019-02-26 2019-09-13 北京知存科技有限公司 一种存算一体芯片中数模转换电路与模数转换电路复用装置

Family Cites Families (25)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6965721B1 (en) * 2001-04-18 2005-11-15 Tullis Barclay J Integrated manufacture of side-polished fiber optics
US20040056852A1 (en) * 2002-09-23 2004-03-25 Jun-Ren Shih Source driver for driver-on-panel systems
US20080303749A1 (en) * 2005-12-13 2008-12-11 Koninklijke Philips Electronics, N.V. Active Matrix Array Device
KR100797751B1 (ko) * 2006-08-04 2008-01-23 리디스 테크놀로지 인코포레이티드 능동 매트릭스 유기 전계 발광 표시 장치의 구동회로
US20080055227A1 (en) * 2006-08-30 2008-03-06 Ati Technologies Inc. Reduced component display driver and method
JP5370714B2 (ja) * 2007-05-31 2013-12-18 ソニー株式会社 光導波路、および信号処理装置
KR100917862B1 (ko) * 2007-12-14 2009-09-18 한국전자통신연구원 다중입력 다중출력 시스템에서 qr 분해 장치 및 그 방법
TW201027315A (en) * 2009-01-13 2010-07-16 Quanta Comp Inc Cell balancing apparatus and operating method thereof
JP5083272B2 (ja) * 2009-05-07 2012-11-28 ソニー株式会社 半導体モジュール
US8681890B2 (en) * 2010-06-07 2014-03-25 Entropic Communications, Inc. Method and apparatus for real time multiplexing with receiver and antenna array elements
US10050680B2 (en) * 2010-06-07 2018-08-14 Entropic Communications, Llc Method and apparatus for real time multiplexing with transmitter and antenna array elements
CN202444479U (zh) 2011-12-15 2012-09-19 北京北广科技股份有限公司 嵌入式立体声音频数模转换器
US8587462B1 (en) * 2012-10-30 2013-11-19 United Microelectronics Corp. Digital-to-analog converter
TWI463155B (zh) * 2012-12-24 2014-12-01 Novatek Microelectronics Corp 具有薄膜覆晶封裝的電子裝置
KR102135635B1 (ko) * 2013-11-20 2020-07-20 엘지디스플레이 주식회사 데이터 구동 집적 회로 및 이를 포함하는 액정 표시 장치
CN203775318U (zh) * 2014-01-17 2014-08-13 中国科学院上海技术物理研究所 基于像素级模数转换的紫外焦平面读出电路
CN204886925U (zh) 2015-06-24 2015-12-16 北京自动测试技术研究所 数模转换器全码测试模块和数模转换器全码测试系统
US9959247B1 (en) * 2017-02-17 2018-05-01 Google Llc Permuting in a matrix-vector processor
US10741088B1 (en) * 2017-09-29 2020-08-11 DroneUp, LLC Multiplexed communications for coordination of piloted aerial drones enlisted to a common mission
CN108763163B (zh) * 2018-08-02 2023-10-20 北京知存科技有限公司 模拟向量-矩阵乘法运算电路
CN111243654A (zh) * 2018-11-28 2020-06-05 北京知存科技有限公司 一种闪存芯片及其校准方法和装置
US11670588B2 (en) * 2019-01-09 2023-06-06 Intel Corporation Selectable vias for back end of line interconnects
CN111614353A (zh) * 2019-02-26 2020-09-01 北京知存科技有限公司 一种存算一体芯片中数模转换电路与模数转换电路复用装置
CN112151095A (zh) * 2019-06-26 2020-12-29 北京知存科技有限公司 存算一体芯片、存储单元阵列结构
CN210780731U (zh) 2019-12-11 2020-06-16 东莞市同门电子科技有限公司 快速稳态数模转换电路及快速稳态数模转换装置

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20040113177A1 (en) * 2002-12-09 2004-06-17 John Wynne Multi-channel integrated circuit comprising a plurality of DACs, and a method for monitoring the output of the DACs
CN104300984A (zh) * 2014-10-21 2015-01-21 上海玮舟微电子科技有限公司 一种模数转换器和模数转换方法
CN108009640A (zh) * 2017-12-25 2018-05-08 清华大学 基于忆阻器的神经网络的训练装置及其训练方法
CN108777155A (zh) * 2018-08-02 2018-11-09 北京知存科技有限公司 闪存芯片
CN208507187U (zh) * 2018-08-02 2019-02-15 北京知存科技有限公司 闪存芯片
CN209390036U (zh) * 2019-02-26 2019-09-13 北京知存科技有限公司 一种存算一体芯片中数模转换电路与模数转换电路复用装置

Also Published As

Publication number Publication date
CN111614353A (zh) 2020-09-01
US20220140834A1 (en) 2022-05-05
US11990915B2 (en) 2024-05-21

Similar Documents

Publication Publication Date Title
WO2020172952A1 (zh) 一种存算一体芯片中数模转换电路与模数转换电路复用装置
US11379673B2 (en) Analog vector-matrix multiplication circuit
WO2021248643A1 (zh) 存算一体电路
US9910827B2 (en) Vector-matrix multiplications involving negative values
US10497442B1 (en) Memristor arrays in crossbars
US11335400B2 (en) Computing-in-memory chip and memory cell array structure
CN109086249B (zh) 模拟向量-矩阵乘法运算电路
WO2021056980A1 (zh) 一种面向卷积神经网络的双相系数可调模拟乘法计算电路
TW200814542A (en) Digital-to-analog converter and image display device
US10027326B2 (en) Reconfigurable circuit
WO2020172951A1 (zh) 可软件定义存算一体芯片及其软件定义方法
CN111949935A (zh) 模拟向量-矩阵乘法运算电路以及芯片
WO2024109644A1 (zh) 忆阻器阵列的操作方法、数据处理装置
CN108599771B (zh) 数模转换电路、方法和显示装置
CN111464764A (zh) 一种基于忆阻器的图像传感器及其进行卷积运算的方法
TW201340620A (zh) 連結至一標準微控制器之作為一整合型週邊設備之類比信號相容互補式金屬氧化物半導體開關
US20180302094A1 (en) Reconfigurable circuit
CN209390036U (zh) 一种存算一体芯片中数模转换电路与模数转换电路复用装置
Sheikholeslami et al. Look-up tables (LUTs) for multiple-valued, combinational logic
CN213365506U (zh) 输出可配置存算一体芯片以及电子设备
KR900003565B1 (ko) 부스(booth) 변환회로
CN112632460A (zh) 源极耦合、漏极求和的模拟向量-矩阵乘法运算电路
US5471161A (en) Circuit for calculating the minimum value
US5467030A (en) Circuit for calculating a maximum value
US6326808B1 (en) Inversion of product term line before or logic in a programmable logic device (PLD)

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 19917161

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205 DATED 02/02/2022)

122 Ep: pct application non-entry in european phase

Ref document number: 19917161

Country of ref document: EP

Kind code of ref document: A1