TWI647627B - Neuromorphic computing system and current estimation method using the same - Google Patents

Abstract

A neurological computing system includes a synapse cell array, a switching circuit, a sensing circuit, and a processing circuit. The synaptic cell array includes a plurality of column lines, a plurality of row lines, and a plurality of synapse units. The synaptic unit is located at the intersection of the column line and the row line. The switching circuit is coupled to the synaptic unit array for electrically connecting the row line to the first terminal or the second terminal. The sensing circuit is coupled to the synaptic cell array for sensing a voltage value and a current value on the row line. The processing circuit is coupled to the switching circuit and the sensing circuit, and configured to: electrically connect a specific one of the plurality of row lines to the first terminal, through the sensing circuit, and electrically connect through the switching circuit The specific row line to the first terminal obtains a first voltage value, passes through the switching circuit, electrically connects the specific row line to the second terminal, transmits the sensing circuit, and electrically connects to the specific line of the second terminal. Obtaining a second voltage value, estimating a product term and a sense current value according to a voltage difference between the first voltage value and the second voltage value.

Description

Neural network computing system and its current estimation method

The present invention generally relates to a neural-like computing system, and more particularly to a neural computing system implemented based on a hardware array structure.

Recently, a neural-like computing device implemented using a hardware array structure has been proposed. A neural-like computing device has the advantage of low power consumption compared to a device that performs a neural-like calculus using a processor (eg, a CPU).

A neurological computing device typically includes a plurality of synapse units. Each synapse unit corresponds to a weight value. When an input vector is applied to the neural network computing device, the input vector is multiplied by the weight vector formed by the weight value corresponding to the associated one or more synaptic units, and a product term sum is formed on the output channel (sum of Product) senses the current. The magnitude of this product and the sense current reflect an integral term and result.

However, as the number of synaptic units increases, the product terms and sense currents on the output channels may become quite large, resulting in increased energy consumption.

The present invention generally relates to a neural computing system implemented based on a hardware array structure. According to an embodiment of the invention, the output channel of the synaptic cell array can be Switchingly connected to the first terminal or the second terminal. The output channel presents a first voltage value when connected to the first terminal and a second voltage value when connected to the second terminal. The product term and the sense current value can be estimated based on the difference between the first voltage value and the second voltage value. Compared with the conventional method, it is possible to directly measure the product flowing through the output channel and the large current to perform an operation. According to the present invention, the output channel connected to the first terminal or the second terminal may only turn on the current whose size is limited, or even It does not conduct current, so it can effectively reduce energy consumption.

According to one aspect of the invention, a neurological computing device is presented. A neurological computing system includes a synaptic cell array, a switching circuit, a sensing circuit, and a processing circuit. The synaptic cell array includes a plurality of column lines, a plurality of row lines, and a plurality of synapse units. The synaptic unit is located at the intersection of the column line and the row line. The switching circuit is coupled to the synapse cell array and configured to connect the row line to the first terminal or the second terminal. The sensing circuit is coupled to the synaptic cell array for sensing a voltage value and a current value on the row line. The processing circuit is coupled to the switching circuit and the sensing circuit, configured to: electrically connect a specific one of the plurality of row lines to the first terminal through the switching circuit, and electrically connect to the first through the sensing circuit The specific row line at the terminal obtains a first voltage value, electrically connects the specific row line to the second terminal through the switching circuit, and obtains the second voltage value through a specific row line when the sensing circuit is electrically connected to the second terminal And estimating the product term and the sense current value according to the voltage difference between the first voltage value and the second voltage value.

According to another aspect of the present invention, a current estimation method for a neural-like computing device is presented. The neurological computing system includes a synaptic cell array, a switching circuit, a sensing circuit, and a processing circuit, and the synaptic cell array includes a plurality of column lines and a plurality of row lines And a plurality of synaptic units located at the intersection of the column lines and the row lines. The current estimation method includes: electrically connecting a specific one of the plurality of row lines to the first terminal through the switching circuit, and obtaining the first voltage value through a specific row line when the sensing circuit is electrically connected to the first terminal And electrically connecting the specific row line to the second terminal through the switching circuit, obtaining the second voltage value through the specific row line when the sensing circuit is electrically connected to the second terminal, and transmitting the processing circuit according to the first voltage value and the second A voltage difference between the voltage values estimates a product term and a sense current value.

In order to better understand the above and other aspects of the present invention, the following detailed description of the embodiments and the accompanying drawings

102‧‧‧ synaptic cell array

104‧‧‧Switching circuit

106‧‧‧Sensor circuit

108‧‧‧Processing circuit

201, 203, 205‧‧‧ column lines

202, 204, 206‧‧‧ lines

210‧‧‧Synaptic unit

WU‧‧‧resistive components

SU‧‧‧Selector

V ₁ , V ₂ , V ₃ ‧‧‧ input voltage

I ₁ , I ₂ , I ₃ ‧‧‧ sense current

T1‧‧‧ first terminal

T2‧‧‧second terminal

SW‧‧‧Switching elements

302, 304, 306, 308, 310‧‧‧ steps

1 is a block diagram of a neurological computing system, schematically depicted in accordance with an embodiment of the present invention.

FIG. 2 schematically shows a circuit configuration diagram of a synapse cell array and a switching circuit.

FIG. 3 is a flow chart of a current estimation method for a neural-like computing system according to an embodiment of the invention.

1 is a block diagram of a neurological computing system, schematically depicted in accordance with an embodiment of the present invention. The neurological computing system includes a synapse cell array 102, a switching circuit 104, a sensing circuit 106, and a processing circuit 108. The switching circuit 104 and the sensing circuit 106 are coupled to the synaptic cell array 102. The processing circuit 108 is coupled to the switching circuit 104 and the sensing circuit 106.

The synapse cell array 102 can multiply the input vector by a weight vector formed by one or more synaptic units to perform a product term sum operation. The switching circuit 104 is controlled by the processing circuit 108 for connecting the various output channels of the synapse cell array 102 to the first terminal or the second terminal. The sensing circuit 106 can sense the voltage value of the output channel as well as the current value. Accordingly, the sensing circuit 106 can obtain a first voltage value presented by the output channel of the synapse cell array 102 when connected to the first terminal and a second voltage value presented when connected to the second terminal. The processing circuit 108 can estimate the magnitude of the product term and the sense current value based on the voltage difference between the first voltage value and the second voltage value.

The sensing circuit 106 includes, for example, a sensing amplifier. Processing circuitry 108 may be implemented, for example, in a microprocessor, microcontroller, wafer, and/or circuit board.

FIG. 2 schematically shows a circuit configuration diagram of the synapse cell array 102 and the switching circuit 104. Although 3 x 3 synaptic units are depicted in FIG. 2, it should be noted that synaptic cell array 102 can include any number of synaptic units and combinations.

As shown in FIG. 2, the synaptic cell array 102 includes a plurality of column lines 201, 203, 205, a plurality of row lines 202, 204, 206, and a plurality of bits at the column lines 201, 203, 205 and the row lines 202, 204, Synaptic unit 210 at the intersection of 206.

Column lines 201,203 synaptic cell array as the input channel 102 to receive an input voltage _{V 1, V 2, V 3} . The input voltages V ₁ , V ₂ , V ₃ can be considered as input vectors [V ₁ , V ₂ , V ₃ ] provided to the system. The synapse unit 210 can respond to the input voltages V ₁ , V ₂ , V ₃ received from the column lines 201 , 203 , 205 , and form a sensing current I ₁ on the row lines 202 , 204 , 206 as output channels of the array, respectively. , I ₂ , I ₃ .

The synapse unit 210 can be any weighting element suitable for a neural-like computing device, such as a "1S1R" circuit structure formed by a series connection of a resistive element WU and a selector SU (eg, a transistor).

The switching circuit 104 can connect one end of each of the row lines 202, 204, 206 to the first terminal T1 or the second terminal T2 through the switching element SW. Both the first terminal T1 and the second terminal T2 are not grounded. Different from the traditional neural computing device, a product term and a large sensing current may be formed on the grounded line, when the row lines 202, 204, 206 are connected to the first terminal T1 or the second terminal T2, on the row lines 202, 204, 206. The sense currents I ₁ , I ₂ , I ₃ will be limited to a preset small current (significantly less than the product term and current), even without current.

In one example, the first terminal T1 is a floating node and the second terminal T2 is a current limiting element such as a current mirror, a transistor or other current source that provides a fixed/limited current.

It should be noted that although three sets of first terminal T1 and second terminal T2 are illustrated in FIG. 2, it should be noted that the switching circuit 104 may include any number of first terminals T1/second terminals T2 and combinations. For example, a plurality of row lines may share the same first terminal T1 and/or second terminal T2.

The sensing circuit 106 is coupled to the row lines 202, 204, 206. The sensing circuit 106 can detect when the row lines 202, 204, 206 are in a first state (ie, a state connected to the first terminal T1) or a second state (ie, a state in which one end of the row line is connected to the second terminal T2) The current value, the voltage value, and the detection result are provided to the processing circuit 108 to estimate the product term and the current value.

For example, assuming that the first terminal T1 is a floating node and the second terminal T2 is a current limiting component, the processing circuit 108 can first set the row to be read (such as the row line 202) by the switching circuit 104. In a state, the first voltage value on the line is obtained by the sensing circuit 106. The processing circuit 108 can then set the sensing circuit 106 to a second state to obtain a second voltage value on the row line and a sense current value that the row line is conducting. In this way, the processing circuit 108 can estimate the product based on the product between the voltage difference between the first voltage value (V _a ) and the second voltage value (V _b ) and the sense current value (I _s ). Item and sense current value (I _sp ). For example, the product term and the sense current value I _sp can be expressed as follows:

Taking the row line 202 as an example, in order to estimate the product term and the current value of the row line 202, the processing circuit 108 may first float the row line 202 through the switching circuit 104 (that is, connect the row line 202 to the first terminal T1). The first voltage value on the row line 202, for example 0.5V, is obtained through the sensing circuit 106.

Next, in response to switching the row line 202 to the second terminal T2 implemented with a 50 μA current source, the processing circuit 108 will pass the sensing circuit 106 to obtain a second voltage value on the row line 202, such as 0.4V.

After obtaining the first voltage value and the second voltage value, the processing circuit 108 can estimate the product of the row line 202 and the sense current value according to (Formula 0) as follows:

To help understand, the following explains why (Equation 0) can be used to estimate the product and current values.

First, it can be seen that when one end of a row line is grounded, the current on the row line is equivalent to the product term and the sense current, which can be expressed as follows:

Where g _out,i represents the weight value of the synaptic unit coupled to the ith column line and the row line to be read, and V _i represents the input voltage value applied to the ith column line, I _out I _ground represents The sense current value (that is, the product term to be estimated and the sense current value (I _sp )) formed when the row line is grounded.

In order to reduce the current generated on the row lines during the product term and operation, the switching circuit 104 can respond to the processing circuit 108 to connect the row line to the floating node (in this example, the first terminal T1). When the row line is floating, the row line will not conduct current (that is, the sense current I _out I _floating on the line is 0), and assumes the balanced voltage value V _out I _floating (in this case, the first voltage) Value V _a ). Therefore, (Formula 1) can be rewritten as follows:

among them .

The switching circuit 104 is further responsive to the processing circuit 108 to connect the row line to the current limiting element (in this example, the second terminal T2). At this time, a sense current I _s will be turned on the line, and has a voltage value V _out I _Iout=Is (in this example, the second voltage value V _b ) is as follows:

Where the value of α is between 0 and 1.

According to (Formula 1), (Formula 2), and (Formula 3), we can obtain:

It can be seen that (Equation 4) has the same mathematical representation as (Equation 0).

In other examples, the first terminal T1 and the second terminal T2 are two current limiting elements corresponding to different current values. At this time, when the one line is terminated by the first terminal T1, the first sensing current is turned on and has a first voltage value. When the row line is terminated by the second terminal T2, the second sense current is turned on and has a second voltage value. By simply modifying the derivation process of (Formula 1) to (Formula 4), the processing circuit 108 can estimate the pair using the first sensed current value, the first voltage value, the second sensed current value, and the second voltage value. The product and current values of the line should be.

FIG. 3 is a flow chart of a current estimation method for a neural-like computing system according to an embodiment of the invention.

In step 302, the switching circuit 104 electrically connects a specific row line to be read to the first terminal T1.

In step 304, the processing circuit 108 transmits the first voltage value through the sensing circuit 106 from the specific row line when electrically connected to the first terminal T1.

At step 306, the switching circuit 104 electrically connects the particular row line to the second terminal T2.

In step 308, the processing circuit 108 transmits the second voltage value through the sensing circuit 106 from the specific row line when electrically connected to the second terminal T2.

At step 310, processing circuit 108 estimates a product term and sense current value based on the voltage difference between the first voltage value and the second voltage value.

In an embodiment, in order to solve the problem that the voltage difference between the first voltage value and the second voltage value is too small to be easily read, the voltage difference can be converted to a time domain through a specially designed sensing technique. To estimate the product term and the sense current value based on the conversion result. For example, the programmable row line charges the capacitor in the first state/second state to obtain a voltage difference between the first voltage value and the second voltage value according to the charge and discharge time of the capacitor, thereby estimating the product term. And sense the current value.

In summary, the present invention generally relates to a neural computing system implemented based on a hardware array structure. According to an embodiment of the invention, the output channel of the synaptic cell array is switchably connected to the first terminal or the second terminal. The output channel presents a first voltage value when connected to the first terminal and a second voltage value when connected to the second terminal. The product term and the sense current value can be estimated based on the difference between the first voltage value and the second voltage value. Compared with the conventional method, it is possible to directly measure the product flowing through the output channel and the large current to perform an operation. According to the present invention, the output channel connected to the first terminal or the second terminal may only turn on the current whose size is limited, or even It does not conduct current, so it can effectively reduce energy consumption.

Although the present invention has been disclosed above by way of example, it is not intended to limit the invention. Those who have ordinary knowledge in the technical field to which the present invention pertains, without departing from the present Within the spirit and scope of the invention, various changes and retouches can be made. Therefore, the scope of the invention is defined by the scope of the appended claims.

Claims (10)

A neurological computing system, comprising: a synaptic cell array, comprising: a plurality of column lines; a plurality of row lines; and a plurality of synaptic units located at intersections of the column lines and the row lines; The circuit is coupled to the synaptic cell array for electrically connecting each of the row lines to a first terminal or a second terminal; a sensing circuit coupled to the synaptic cell array for sensing the a voltage value and a current value on the line; and a processing circuit coupled to the switching circuit and the sensing circuit, and configured to: pass the switching circuit to select a specific one of the row lines Connected to the first terminal; the first voltage value is obtained by the sensing circuit, the specific row line when electrically connected to the first terminal; and the specific row line is electrically connected to the a second terminal; the second voltage value is obtained by the sensing circuit, the specific row line when electrically connected to the second terminal; and a voltage between the first voltage value and the second voltage value The difference is estimated by an integral term and a sensed current value. The neural computing system of claim 1, wherein the first terminal is a floating node, and the second terminal is a current limiting component. The neural computing system of claim 2, wherein the processing circuit is further configured to obtain a sensing current value on the specific row line from the sensing circuit when the specific row line is electrically connected to the second terminal. And estimating the product term and the sense current value based on the product of the voltage difference and the sensed current value. For example, the neural computing system of claim 3, wherein the product and the sensed current value (I _sp ) are: Where I _s is the sense current value, V _a is the first voltage value, and V _b is the second voltage value. A neural computing system as in claim 2, wherein the current limiting element is a current mirror or a transistor. A current estimation method for a neural computing system, the neural computing system comprising a synapse cell array, a switching circuit, a sensing circuit and a processing circuit, the synaptic cell array comprising a plurality of column lines and a plurality of row lines And a plurality of synaptic units located at the intersection of the column lines and the row lines, the current estimation method includes: electrically connecting a specific one of the plurality of row lines to the first through the switching circuit terminal; Transmitting, by the switching circuit, the specific row line is electrically connected to the second terminal through the sensing circuit, and the sensing circuit is configured to electrically connect the specific row line to the second terminal; a second voltage value obtained by electrically connecting the second line to the second terminal; and transmitting, by the processing circuit, a voltage difference between the first voltage value and the second voltage value, Estimate a product and sense current value. The current estimation method of claim 6, wherein the first terminal is a floating node, and the second terminal is a current limiting component. The current estimation method of claim 7, further comprising: when the specific row line is electrically connected to the second terminal, using the sensing circuit to obtain a sensing current value on the specific row line; and using the processing The circuit estimates the product term and the sense current value based on the product of the voltage difference and the sensed current value. For example, the current estimation method of claim 8 wherein the product term and the sense current value (I _sp ) are: Where I _s is the sense current value, V _a is the first voltage value, and V _b is the second voltage value. The current estimation method of claim 7, wherein the current limiting element is a current mirror or a transistor.