Artificial neural network including three-dimensional integrated circuit
By using nonvolatile memory cell arrays in neural networks to achieve independent programming and continuous programming, the problems of low energy efficiency and high computational complexity of neural networks in the prior art are solved, and efficient simulated computing and fine synapses are realized Tune.
Patent Information
- Application Number
- CN202280093793.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Priority Date
- 2022-06-23
- Filing Date
- 2022-07-19
- Publication Date
- 2024-11-08
AI Technical Summary
The prior art lacks sufficient hardware technology when developing artificial neural networks with high performance information processing, resulting in low energy efficiency and high computational complexity of neural networks. The synapses achieved by CMOS analog circuits are too large, making it difficult to achieve efficient neural networks. Connection between elements.
Using a nonvolatile memory cell array as synapses of neural networks, the continuous programming is achieved by configuring each memory cell to independently program, erase and read, allowing multiple discrete values to be stored, thereby achieving fine synaptic weight tuning .
The energy efficiency and calculation efficiency of neural networks are improved, the demand for separate multiplication and addition logic circuits is reduced, and high-efficiency analog calculation is achieved.
Smart Images

Figure CN118922840A_ABST
Abstract
Description
[0001] Priority declaration
[0002] This application claims priority to U.S. Provisional Patent Application No. 63 / 328,126, filed on April 6, 2022, entitled “Artificial Neural Network Comprising a Three-Dimensional Integrated Circuit,” and U.S. Patent Application No. 17 / 848,371, filed on June 23, 2022, entitled “Artificial Neural Network Comprising a Three-Dimensional Integrated Circuit.” Technical Field
[0003] Many examples of an artificial neural network including a three-dimensional integrated circuit are disclosed. Background Art
[0004] Artificial neural networks simulate biological neural networks (the central nervous system of animals, especially the brain), and are used to estimate or approximate functions that may depend on a large number of inputs and are usually unknown. Artificial neural networks typically consist of layers of interconnected "neurons" that exchange messages with each other.
[0005] FIG1 shows an artificial neural network, where circles represent inputs or layers of neurons. Connections (called synapses) are represented by arrows and have numerical weights that can be tuned based on experience. This allows the neural network to adapt to the input and be able to learn. Typically, a neural network includes multiple layers of inputs. There are typically one or more intermediate layers of neurons, and an output layer of neurons that provide the output of the neural network. Neurons at each level make decisions based on data received from the synapses, either individually or collectively.
[0006] One of the main challenges in developing artificial neural networks for high-performance information processing is the lack of adequate hardware technology. In fact, practical neural networks rely on a large number of synapses to achieve high connectivity between neurons, i.e. very high computational parallelism. In principle, this complexity can be achieved by digital supercomputers or clusters of dedicated graphics processing units. However, these approaches, in addition to their high cost, are also mediocre in energy efficiency compared to biological networks, which consume much less energy mainly due to the low-precision analog calculations they perform. CMOS analog circuits have been used for artificial neural networks, but given the large number of neurons and synapses, the synapses of most CMOS implementations are too large.
[0007] Applicant previously disclosed an artificial (simulated) neural network using one or more non-volatile memory arrays as synapses in U.S. Patent Application Publication 2017 / 0337466A1, which is incorporated herein by reference. The non-volatile memory array operates as an analog neural memory and includes non-volatile memory cells arranged in rows and columns. The neural network includes a first plurality of synapses configured to receive a first plurality of inputs and generate a first plurality of outputs therefrom, and a first plurality of neurons configured to receive a first plurality of outputs. The first plurality of synapses includes a plurality of memory cells, wherein each of the memory cells includes: a source region and a drain region spaced apart formed in a semiconductor substrate, wherein a channel region extends between the source region and the drain region; a floating gate disposed above a first portion of the channel region and insulated from the first portion; and a non-floating gate disposed above a second portion of the channel region and insulated from the second portion. Each of the plurality of memory cells stores a weight value corresponding to a plurality of electrons on the floating gate. The plurality of memory cells multiply the first plurality of inputs by the stored weight values to generate a first plurality of outputs.
[0008] Non-volatile memory cell
[0009] Nonvolatile memory is well known. For example, U.S. Pat. No. 5,029,130 (“the '130 patent”), which is incorporated herein by reference, discloses an array of split-gate nonvolatile memory cells, which is a type of flash memory cell. Such a memory cell 210 is shown in FIG. 2 . Each memory cell 210 includes a source region 14 and a drain region 16 formed in a semiconductor substrate 12 with a channel region 18 therebetween. A floating gate 20 is formed over and insulated from (and controls the conductivity of) a first portion of the channel region 18, and is formed over a portion of the source region 14. A word line terminal 22 (which is typically coupled to a word line) has a first portion disposed over and insulated from (and controls the conductivity of) a second portion of the channel region 18, and a second portion extending upward and located over the floating gate 20. The floating gate 20 and the word line terminal 22 are insulated from the substrate 12 by a gate oxide. A bit line 24 is coupled to the drain region 16.
[0010] Memory cell 210 is erased (where electrons are removed from the floating gate) by placing a high positive voltage on word line terminal 22, which causes the electrons on floating gate 20 to tunnel through the intervening insulator via Fowler-Nordheim (FN) tunneling from floating gate 20 to word line terminal 22.
[0011] The memory cell 210 is programmed by source side injection (SSI) with hot electrons (where electrons are placed on the floating gate) by placing a positive voltage on the word line terminal 22 and a positive voltage on the source region 14. The electron flow will flow from the drain region 16 to the source region 14. When the electrons reach the gap between the word line terminal 22 and the floating gate 20, the electrons will accelerate and become hot. Due to the electrostatic attraction from the floating gate 20, some of the heated electrons will be injected through the gate oxide onto the floating gate 20.
[0012] The memory cell 210 is read by placing a positive read voltage on the drain region 16 and the word line terminal 22 (which turns on the portion of the channel region 18 below the word line terminal). If the floating gate 20 is positively charged (i.e., the electrons are erased), the portion of the channel region 18 below the floating gate 20 is also turned on, and current will flow through the channel region 18, which is sensed as an erased state or a "1" state. If the floating gate 20 is negatively charged (i.e., programmed by electrons), the portion of the channel region below the floating gate 20 is mostly or completely turned off, and current will not (or very little) flow through the channel region 18, which is sensed as a programmed state or a "0" state.
[0013] Table 1 depicts typical voltage and current ranges that may be applied to the terminals of the memory cell 210 for performing read operations, erase operations, and program operations:
[0014] Table 1: Operation of the flash memory unit 210 of FIG. 2
[0015] WL BL SL Read 2V-3V 0.6V-2V 0V Erase About 11V-13V 0V 0V programming 1-2V 10.5μA-3μA 9V-10V
[0016] Other split gate memory cell configurations are known as other types of flash memory cells. For example, FIG. 3 depicts a four-gate memory cell 310, which includes a source region 14, a drain region 16, a floating gate 20 above a first portion of a channel region 18, a select gate 22 (typically coupled to a word line WL) above a second portion of the channel region 18, a control gate 28 above the floating gate 20, and an erase gate 30 above the source region 14. This configuration is described in U.S. Pat. No. 6,747,310, which is incorporated herein by reference for all purposes. Here, except for the floating gate 20, all gates are non-floating gates, meaning that they are electrically connected or capable of being electrically connected to a voltage source. Programming is performed by heated electrons from the channel region 18 that inject themselves into the floating gate 20. Erasing is performed by electrons tunneling from the floating gate 20 to the erase gate 30.
[0017] Table 2 depicts typical voltage and current ranges that may be applied to the terminals of the memory cell 310 for performing read operations, erase operations, and program operations:
[0018] Table 2: Operation of the flash memory unit 310 of FIG. 3
[0019] WL / SG BL CG EG SL Read 1.0V-2V 0.6V-2V 0V-2.6V 0V-2.6V 0V Erase -0.5V / 0V 0V 0V / -8V 8-12V 0V programming 1V 0.1μA-1μA 8V-11V 4.5V-9V 4.5V-5V
[0020] FIG4 depicts a tri-gate memory cell 410, which is another type of flash memory cell. Memory cell 410 is the same as memory cell 310 of FIG3, except that memory cell 410 does not have a separate control gate. Erase operations (thus erasing by using an erase gate) and read operations are similar to the operations of FIG3, except that no control gate bias is applied. Program operations are also completed without a control gate bias, and therefore, a higher voltage is applied on the source line during a programming operation to compensate for the lack of a control gate bias.
[0021] Table 3 depicts typical voltage and current ranges that may be applied to the terminals of the memory cell 410 for performing read operations, erase operations, and program operations:
[0022] Table 3: Operation of the flash memory unit 410 of FIG. 4
[0023] WL / SG BL EG SL Read 0.7V-2.2V 0.6V-2V 0V-2.6V 0V Erase -0.5V / 0V 0V 11.5V 0V programming 1V 0.2μA-3μA 4.5V 7V-9V
[0024] FIG5 depicts a stacked gate memory cell 510, which is another type of flash memory cell. Memory cell 510 is similar to memory cell 210 of FIG2, except that floating gate 20 extends over the entire channel region 18, and control gate 22 (which here will be coupled to the word line) extends over floating gate 20, separated by an insulating layer (not shown). Erasing is performed by FN tunneling of electrons from the FG to the substrate, programming is performed by channel hot electron (CHE) injection at the region between channel 18 and drain region 16, by electron flow from source region 14 toward drain region 16, and a read operation similar to memory cell 210 with a higher control gate voltage.
[0025] Table 4 depicts typical voltage ranges that may be applied to the terminals of the memory cell 510 and the substrate 12 for performing read operations, erase operations, and program operations:
[0026] Table 4: Operation of the flash memory unit 510 of FIG. 5
[0027] CG BL SL Substrate Read 2-5V 0.6-2V 0V 0V Erase -8V to -10V / 0V FLT FLT 8-10V / 15-20V programming 8-12V 3-5V 0V 0V
[0028] The methods and devices described herein can be applied to other non-volatile memory technologies such as, but not limited to, FINFET split-gate flash or stacked-gate flash memory, NAND flash, SONOS (silicon-oxide-nitride-oxide-silicon, charge trapped in nitride), MONOS (metal-oxide-nitride-oxide-silicon, metal charge trapped in nitride), ReRAM (resistive random access memory), PCM (phase change memory), MRAM (magnetic ram), FeRAM (ferroelectric ram), CT (charge trapping) memory, CN (carbon tube) memory, OTP (two-level or multi-level one-time programmable) and CeRAM (correlated electron ram), etc.
[0029] In order to utilize a memory array comprising one of the above-described types of non-volatile memory cells in an artificial neural network, two modifications are made. First, the circuitry is configured so that each memory cell can be individually programmed, erased, and read without adversely affecting the memory states of other memory cells in the array, as further explained below. Second, continuous (analog) programming of the memory cells is provided.
[0030] Specifically, the memory state (i.e., the charge on the floating gate) of each memory cell in the array can be changed from a fully erased state to a fully programmed state, and vice versa, independently and continuously with minimal disturbance to other memory cells. This means that the cell storage device is actually analog, or at least can store one of many discrete values (such as 16 or 64 different values), which allows very precise and individual tuning of all the memory cells in the memory array, and makes the memory array ideal for storing and fine tuning the synaptic weights of a neural network.
[0031] Neural Network Using Non-Volatile Memory Cell Array
[0032] Figure 6 A non-limiting example of a neural network utilizing a non-volatile memory array of this example is conceptually illustrated. This example uses a non-volatile memory array neural network for a facial recognition application, but any other suitable application may also be implemented using a neural network based on a non-volatile memory array.
[0033] For this example, S0 is the input layer, which is a 32×32 pixel RGB image with 5-bit precision (i.e., three 32×32 pixel arrays, one for each color R, G, and B, with 5-bit precision per pixel). Synapse CB1 from input layer S0 to layer C1 applies different sets of weights in some cases and shared weights in other cases, and scans the input image with a 3×3 pixel overlapping filter (kernel), shifting the filter by 1 pixel (or more than 1 pixel as the model dictates). Specifically, the values of 9 pixels in the 3×3 portion of the image (i.e., called the filter or kernel) are provided to synapse CB1, where these 9 input values are multiplied by the appropriate weights, and after summing the outputs of the multiplications, a single output value is determined and provided by the first synapse of CB1 for generating a pixel of one of the feature maps of layer C1. The 3×3 filter is then shifted right by one pixel within the input layer S0 (i.e., a column of three pixels on the right is added, and a column of three pixels on the left is released), whereby the 9 pixel values in this newly positioned filter are provided to the synapse CB1, where they are multiplied by the same weights and a second single output value is determined by the associated synapse. This process continues until the 3×3 filter has scanned all three colors and all bits (precision values) over the entire 32×32 pixel image of the input layer S0. The process is then repeated using different sets of weights to generate different feature maps for layer C1 until all feature maps for layer C1 are calculated.
[0034] At layer C1, in this example, there are 16 feature maps, each with 30×30 pixels. Each pixel is a new feature pixel extracted from the product of the input and the kernel, so each feature map is a two-dimensional array, so in this example, layer C1 consists of a two-dimensional array of 16 layers (remember that the layers and arrays referred to in this article are logical relationships, not necessarily physical relationships, that is, arrays do not have to be oriented to physical two-dimensional arrays). Each of the 16 feature maps in layer C1 is generated by one of sixteen different sets of synaptic weights applied to the filter scan. The C1 feature maps can all relate to different aspects of the same image features, such as boundary recognition. For example, a first map (generated using a first set of weights, shared for all scans used to generate the first map) can identify circular edges, a second map (generated using a second set of weights different from the first set of weights) can identify rectangular edges, or the aspect ratio of certain features, and so on.
[0035] Before passing from layer C1 to layer S1, an activation function P1 (pooling) is applied, which pools the values from consecutive non-overlapping 2×2 regions in each feature map. The purpose of the pooling function P1 is to average neighboring positions (or a max function can also be used) to, for example, reduce the dependency of edge positions and reduce the data size before entering the next stage. At layer S1, there are 16 15×15 feature maps (i.e., sixteen different arrays of 15×15 pixels per feature map). The synapse CB2 from layer S1 to layer C2 scans the map in layer S1 using a 4×4 filter, where the filter is shifted by 1 pixel. At layer C2, there are 22 12×12 feature maps. Before passing from layer C2 to layer S2, an activation function P2 (pooling) is applied, which pools the values from consecutive non-overlapping 2×2 regions in each feature map. At layer S2, there are 22 6×6 feature maps. An activation function (pooling) is applied to the synapse CB3 from layer S2 to layer C3, where each neuron in layer C3 is connected to each map in layer S2 via a corresponding synapse of CB3. At layer C3, there are 64 neurons. Synapse CB4 from layer C3 to output layer S3 completely connects C3 to S3, i.e., each neuron in layer C3 is connected to each neuron in layer S3. The output at S3 includes 10 neurons, where the highest output neuron determines the class. For example, the output may indicate recognition or classification of the content of the original image.
[0036] The synapses at each layer are implemented using an array or a portion of an array of non-volatile memory cells.
[0037] Figure 7 A block diagram of an array that can be used for this purpose is shown in FIG. 3. The vector-matrix multiplication (VMM) array 32 includes non-volatile memory cells and serves as a synapse between one layer and the next (such as Figure 6 CB1, CB2, CB3 and CB4 in FIG. 3 ). Specifically, the VMM array 32 includes a nonvolatile memory cell array 33, an erase gate and word line gate decoder 34, a control gate decoder 35, a bit line decoder 36 and a source line decoder 37, which decode the corresponding inputs of the nonvolatile memory cell array 33. The input to the VMM array 32 may come from the erase gate and word line gate decoder 34 or from the control gate decoder 35. In this example, the source line decoder 37 also decodes the output of the nonvolatile memory cell array 33. Alternatively, the bit line decoder 36 may decode the output of the nonvolatile memory cell array 33.
[0038] The non-volatile memory cell array 33 serves two purposes. First, it stores weights to be used by the VMM array 32. Second, the non-volatile memory cell array 33 effectively multiplies the input with the weights stored in the non-volatile memory cell array 33 and each output line (source line or bit line) adds them to produce an output, which will serve as the input of the next layer or the input of the final layer. By performing multiplication and addition functions, the non-volatile memory cell array 33 eliminates the need for separate multiplication and addition logic circuits, and is also power efficient due to its in-situ memory calculations.
[0039] The output of the non-volatile memory cell array 33 is provided to a differential summer (such as a summing operational amplifier or a summing current mirror) 38, which sums the output of the non-volatile memory cell array 33 to create a single value for the convolution. The differential summer 38 is arranged to perform the summing of positive and negative weights.
[0040] The summed output value of the difference summer 38 is then provided to the activation function block 39, which modifies the output. The activation function block 39 may provide a sigmoid, tanh, or ReLU function. The modified output value of the activation function block 39 becomes the next layer (e.g., Figure 6 The elements of the feature map of layer C1 in the image are then applied to the next synapse to produce the next feature map layer or the final layer. Thus, in this example, the non-volatile memory cell array 33 constitutes a plurality of synapses (which receive their inputs from an existing neuron layer or from an input layer such as an image database), and the summing operational amplifier 38 and the activation function block 39 constitute a plurality of neurons.
[0041] Figure 7 The inputs to the VMM array 32 (WLx, EGx, CGx, and optionally BLx and SLx) may be analog levels, binary levels, or digital bits (in which case a DAC is provided to convert the digital bits to the appropriate input analog levels), and the outputs may be analog levels, binary levels, or digital bits (in which case an output ADC is provided to convert the output analog levels to digital bits).
[0042] Figure 8 1 is a block diagram depicting the use of multiple layers of VMM arrays 32, labeled here as VMM arrays 32a, 32b, 32c, 32d, and 32e. Figure 8As shown, the input (denoted as Inputx) is converted from digital to analog by a digital-to-analog converter 31 and provided to an input VMM array 32a. The converted analog input may be a voltage or a current. The first level of input D / A conversion may be accomplished by using a function or LUT (lookup table) that maps the input Inputx to the appropriate analog level of the matrix multiplier of the input VMM array 32a. The input conversion may also be accomplished by an analog-to-analog (A / A) converter to convert an external analog input into a mapped analog input to the input VMM array 32a.
[0043] The output generated by the input VMM array 32a is provided as input to the next VMM array (hidden level 1) 32b, which in turn generates outputs provided as input to the next VMM array (hidden level 2) 32c, and so on. The layers of the VMM array 32 serve as different layers of synapses and neurons of a convolutional neural network (CNN). Each VMM array 32a, 32b, 32c, 32d, and 32e can be an independent physical non-volatile memory array, or multiple VMM arrays can utilize different portions of the same non-volatile memory array, or multiple VMM arrays can utilize overlapping portions of the same physical non-volatile memory array. Figure 8 The example shown includes five layers (32a, 32b, 32c, 32d, 32e): one input layer (32a), two hidden layers (32b, 32c) and two fully connected layers (32d, 32e). One of ordinary skill in the art will appreciate that this is merely exemplary and that, on the contrary, the system may include more than two hidden layers and more than two fully connected layers.
[0044] Vector-Matrix Multiplication (VMM) Array
[0045] Fig. 9 Depicted is a neuron VMM array 900 that is particularly suitable for use with the memory cells 310 shown in FIG. 3 and serves as a synapse and component of neurons between an input layer and the next layer. The VMM array 900 includes a memory array 901 of nonvolatile memory cells and a reference array 902 of nonvolatile reference memory cells (at the top of the array). Alternatively, another reference array may be placed at the bottom.
[0046] In the VMM array 900, control gate lines (such as control gate line 903) extend in the vertical direction (so reference array 902 is orthogonal to control gate line 903 in the row direction), and erase gate lines (such as erase gate line 904) extend in the horizontal direction. Here, the inputs to the VMM array 900 are set on the control gate lines (CG0, CG1, CG2, CG3), and the outputs of the VMM array 900 appear on the source lines (SL0, SL1). In one example, only even rows are used, and in another example, only odd rows are used. The current placed on each source line (SL0, SL1, respectively) performs a summing function of all currents from the memory cells connected to that particular source line.
[0047] As described herein for neural networks, non-volatile memory cells of VMM array 900 (ie, memory cells 310 of VMM array 900 ) may be configured to operate in a sub-threshold region.
[0048] The nonvolatile reference memory cell and the nonvolatile memory cell described herein are biased in weak inversion (subthreshold region):
[0049] Ids=Io*e (Vg-Vth) / nVt =w*Io*e (Vg) / nVt ,
[0050] where w = e (-Vth) / nVt
[0051] Where Ids is the drain to source current; Vg is the gate voltage on the memory cell; Vth is the threshold voltage of the memory cell; Vt is the thermal voltage = k*T / q, where k is the Boltzmann constant, T is the temperature in Kelvin, and q is the electron charge; n is the slope factor = 1+(Cdep / Cox), where Cdep = the capacitance of the depletion layer, and Cox is the capacitance of the gate oxide layer; Io is the memory cell current at a gate voltage equal to the threshold voltage, and Io is related to (Wt / L)*u*Cox*(n-1)*Vt 2 is proportional to , where u is the carrier mobility, and Wt and L are the width and length of the memory cell, respectively.
[0052] For an I to V logarithmic converter that uses a memory cell (such as a reference memory cell or a peripheral memory cell) or a transistor to convert the input current into an input voltage:
[0053] Vg=n*Vt*log[Ids / wp*Io]
[0054] Wherein, wp is w of a reference memory cell or a peripheral memory cell.
[0055] For a memory array used as a vector matrix multiplier VMM array with current input, the output current is:
[0056] Iout=wa*Io*e (Vg) / nVt ,Right now
[0057] Iout=(wa / wp)*Iin=W*Iin
[0058] W=e (Vthp-Vtha) / nVt
[0059] Here, wa = w for each memory cell in the memory array.
[0060] Vthp is the effective threshold voltage of the peripheral memory cells, and Vtha is the effective threshold voltage of the main (data) memory cells. Note that the threshold voltage of the transistor is a function of the substrate bulk bias voltage, and the substrate bulk bias voltage, denoted as Vsb, can be modulated to compensate for various conditions at this temperature. The threshold voltage Vth can be expressed as:
[0061] Vth=Vth0+γ(SQRT|Vsb-2*φF)-SQRT|2*φF|)
[0062] where Vth0 is the threshold voltage with zero substrate bias, φF is the surface potential, and γ is the bulk effect parameter.
[0063] The word line or control gate may be used as the input of the memory cell for the input voltage.
[0064] Alternatively, the flash memory cells of the VMM array described herein may be configured to operate in the linear region:
[0065] Ids=β*(Vgs-Vth)*Vds;β=u*Cox*Wt / L
[0066] W=α(Vgs-Vth)
[0067] This means that the weight W in the linear region is proportional to (Vgs-Vth)
[0068] The word line or control gate or bit line or source line can be used as the input of the memory cell operating in the linear region. The bit line or source line can be used as the output of the memory cell.
[0069] For an IV linear converter, a memory cell (such as a reference memory cell or a peripheral memory cell) or a transistor operating in a linear region may be used to linearly convert an input / output current into an input / output voltage.
[0070] Alternatively, the memory cells of the VMM array described herein may be configured to operate in the saturation region:
[0071] Ids=1 / 2*β*(Vgs-Vth) 2 ; β=u*Cox*Wt / L
[0072] Wα(Vgs-Vth) 2 , which means weight W and (Vgs-Vth) 2 Proportional
[0073] The word line, control gate, or erase gate can be used as the input of the memory cell operating in the saturation region. The bit line or source line can be used as the output of the output neuron.
[0074] Alternatively, the memory cells of the VMM array described herein may be used for all regions or combinations thereof (subthreshold, linear or saturation regions) of each layer or multiple layers of a neural network.
[0075] U.S. Patent No. 10,748,630 describes Figure 7 Other examples of VMM arrays 32 of , which are incorporated herein by reference. As described herein, source lines or bit lines can be used as neuron outputs (current summing outputs).
[0076] Fig.10 A neuron VMM array 1000 is depicted, which is particularly suitable for the memory cell 210 shown in FIG. 2 and serves as a synapse between an input layer and a next layer. The VMM array 1000 includes a memory array 1003 of nonvolatile memory cells, a reference array 1001 of a first nonvolatile reference memory cell, and a reference array 1002 of a second nonvolatile reference memory cell. The reference arrays 1001 and 1002 arranged in the column direction of the array are used to convert the current input flowing into the terminals BLR0, BLR1, BLR2, and BLR3 into voltage inputs WL0, WL1, WL2, and WL3. In practice, the first nonvolatile reference memory cell and the second nonvolatile reference memory cell are diode-connected through a multiplexer 1014 (only partially depicted), in which the current input flows. The reference cells are tuned (e.g., programmed) to a target reference level. The target reference level is provided by a reference microarray matrix (not shown).
[0077] Memory array 1003 serves two purposes. First, it stores the weights that will be used by VMM array 1000 on their corresponding memory cells. Second, memory array 1003 effectively converts the inputs (i.e., the current inputs provided at terminals BLR0, BLR1, BLR2, and BLR3) that reference arrays 1001 and 1002 convert into input voltages to provide to the VMM array 1000.
[0078] The word lines WL0, WL1, WL2, and WL3) are multiplied by the weights stored in the memory array 1003, and all the results (memory cell currents) are then added to produce an output on the corresponding bit lines (BL0-BLN), which will be the input to the next layer or the input to the final layer. By performing multiplication and addition functions, the memory array 1003 eliminates the need for separate multiplication logic circuits and addition logic circuits, and is also highly power-efficient. Here, voltage inputs are provided on word lines WL0, WL1, WL2, and WL3, and outputs appear on the corresponding bit lines BL0-BLN during a read (inference) operation. The current placed on each of the bit lines BL0-BLN performs a summing function of the currents from all non-volatile memory cells connected to that particular bit line.
[0079] Table 5 depicts operating voltages and currents for the VMM array 1000. The columns in the table indicate the voltages placed on the word line for a selected cell, the word line for an unselected cell, the bit line for a selected cell, the bit line for an unselected cell, the source line for a selected cell, and the source line for an unselected cell. The rows indicate read, erase, and program operations.
[0080] Table 5: Fig.10 Operation of VMM Array 1000 :
[0081] WL WL-Not selected BL BL-Not selected SL SL-Not selected Read 1V-3.5V -0.5V / 0V 0.6-2V(Ineuron) 0.6V-2V / 0V 0V 0V Erase About 5-13V 0V 0V 0V 0V 0V programming 1-2V -0.5V / 0V 0.1-3uA Vinh~2.5V 4-10V 0V-1V / FLT
[0082] Fig.11 A neuron VMM array 1100 is depicted, which is particularly suitable for the memory cell 210 shown in FIG. 2 and serves as a synapse and component of a neuron between an input layer and a next layer. The VMM array 1100 includes a memory array 1103 of nonvolatile memory cells, a reference array 1101 of a first nonvolatile reference memory cell, and a reference array 1102 of a second nonvolatile reference memory cell. The reference arrays 1101 and 1102 extend in the row direction of the VMM array 1100. The VMM array is similar to the VMM 1000, except that in the VMM array 1100, the word lines extend in the vertical direction. Here, the inputs are set on the word lines (WLA0, WLB0, WLA1, WLB2, WLA2, WLB2, WLA3, WLB3), and the outputs appear on the source lines (SL0, SL1) during a read operation. The current placed on each source line performs a summing function of all currents from the memory cells connected to that particular source line.
[0083] Table 6 depicts operating voltages and currents for the VMM array 1100. The columns in the table indicate the voltages placed on the word line for a selected cell, the word line for an unselected cell, the bit line for a selected cell, the bit line for an unselected cell, the source line for a selected cell, and the source line for an unselected cell. The rows indicate read, erase, and program operations.
[0084] Table 6: Fig.11 Operation of the VMM array 1100
[0085]
[0086] Fig.12 A neuron VMM array 1200 is depicted, which is particularly suitable for the memory cell 310 shown in FIG. 3 and is used as a synapse and component of a neuron between an input layer and a next layer. The VMM array 1200 includes a memory array 1203 of nonvolatile memory cells, a reference array 1201 of a first nonvolatile reference memory cell, and a reference array 1202 of a second nonvolatile reference memory cell. The reference arrays 1201 and 1202 are used to convert the current input flowing into the terminals BLR0, BLR1, BLR2, and BLR3 into voltage inputs CG0, CG1, CG2, and CG3. In practice, the first nonvolatile reference memory cell and the second nonvolatile reference memory cell are diode-connected through a multiplexer 1212 (only partially shown), wherein the current input flows into them through BLR0, BLR1, BLR2, and BLR3. Multiplexers 1212 each include a respective multiplexer 1205 and cascode transistor 1204 to ensure a constant voltage on a bit line (such as BLR0) of each of the first and second nonvolatile reference memory cells during a read operation. The reference cells are tuned to a target reference level.
[0087] The memory array 1203 serves two purposes. First, it stores weights to be used by the VMM array 1200. Second, the memory array 1203 effectively multiplies the inputs (current inputs provided to terminals BLR0, BLR1, BLR2, and BLR3, which are converted by reference arrays 1201 and 1202 into input voltages to be provided to the control gates (CG0, CG1, CG2, and CG3)) by the weights stored in the memory array, and then adds all the results (cell currents) to produce an output, which appears in BL0-BLN and will be the input to the next layer or the input to the final layer. By performing multiplication and addition functions, the memory array eliminates the need for separate multiplication and addition logic circuits and is also highly power efficient. Here, the inputs are provided on the control gate lines (CGO, CG1, CG2, and CG3), and the outputs appear on the bit lines (BL0-BLN) during a read operation. The current placed on each bit line performs a summing function of all currents from the memory cells connected to that particular bit line.
[0088] The VMM array 1200 implements unidirectional tuning of the non-volatile memory cells in the memory array 1203. That is, each non-volatile memory cell is erased and then partially programmed until the desired charge on the floating gate is reached. If too much charge is placed on the floating gate (causing an erroneous value to be stored in the cell), the cell is erased and the sequence of partial programming operations restarts. As shown, two rows sharing the same erase gate (such as EG0 or EG1) are erased together (which is called a page erase), and thereafter, each cell is partially programmed until the desired charge on the floating gate is reached.
[0089] Table 7 depicts operating voltages and currents for the VMM array 1200. The columns in the table indicate the voltages placed on the word line for a selected cell, the word line for an unselected cell, the bit line for a selected cell, the bit line for an unselected cell, the control gate for a selected cell, the control gate for an unselected cell in the same sector as the selected cell, the control gate for an unselected cell in a different sector than the selected cell, the erase gate for a selected cell, the erase gate for an unselected cell, the source line for a selected cell, and the source line for an unselected cell. The rows indicate read, erase, and program operations.
[0090] Table 7: Fig.12 Operation of the VMM array 1200
[0091]
[0092] Fig.13A neuron VMM array 1300 is depicted that is particularly suitable for use with the memory cell 310 shown in FIG. 3 and serves as a synapse and component of a neuron between an input layer and a next layer. The VMM array 1300 includes a memory array 1303 of nonvolatile memory cells, a reference array 1301 of a first nonvolatile reference memory cell, and a reference array 1302 of a second nonvolatile reference memory cell. The EG lines EGR0, EG0, EG1, and EGR1 extend vertically, while the CG lines CG0, CG1, CG2, and CG3 and the SL lines WL0, WL1, WL2, and WL3 extend horizontally. The VMM array 1300 is similar to the VMM array 1400, except that the VMM array 1300 implements bidirectional tuning, where each individual cell can be fully erased, partially programmed, and partially erased as needed to achieve a desired amount of charge on the floating gate due to the use of separate EG lines. As shown, reference arrays 1301 and 1302 convert input currents in terminals BLR0, BLR1, BLR2, and BLR3 into control gate voltages CG0, CG1, CG2, and CG3 to be applied to the memory cells in the row direction (through the action of diode-connected reference cells via multiplexer 1314). The current outputs (neurons) are in bit lines BL0-BLN, where each bit line sums all currents from the non-volatile memory cells connected to that particular bit line.
[0093] Table 8 depicts operating voltages and currents for the VMM array 1300. The columns in the table indicate the voltages placed on the word line for a selected cell, the word line for an unselected cell, the bit line for a selected cell, the bit line for an unselected cell, the control gate for a selected cell, the control gate for an unselected cell in the same sector as the selected cell, the control gate for an unselected cell in a different sector than the selected cell, the erase gate for a selected cell, the erase gate for an unselected cell, the source line for a selected cell, and the source line for an unselected cell. The rows indicate read, erase, and program operations.
[0094] Table 8: Fig.13 Operation of the VMM array 1300
[0095]
[0096] Fig. 22 A neuron VMM array 2200 is depicted, which is particularly suitable for the memory unit 210 shown in FIG. 2 and is used as a synapse and component of neurons between an input layer and the next layer. In the VMM array 2200, inputs INPUT0, ..., INPUT N On bit lines BL0, BL NThe signals are received on the source lines SL0, SL1, SL2 and SL3, and outputs OUTPUT1, OUTPUT2, OUTPUT3 and OUTPUT4 are generated on the source lines SL0, SL1, SL2 and SL3, respectively.
[0097] Fig.23 A neuron VMM array 2300 is depicted that is particularly suitable for use with the memory cell 210 shown in FIG. 2 and serves as a synapse and component of neurons between an input layer and the next layer. In this example, inputs INPUT0, INPUT1, INPUT2, and INPUT3 are received on source lines SL0, SL1, SL2, and SL3, respectively, and outputs OUTPUT0, ..., OUTPUT N On the bit lines BL0, BL N Generate on.
[0098] Fig.24 A neuron VMM array 2400 is depicted, which is particularly suitable for the memory unit 210 shown in Figure 2 and is used as a synapse and component of neurons between the input layer and the next layer. In this example, input INPUT0, ..., INPUT M On word lines WL0, ......, WL M is received and output OUTPUT0, ......, OUTPUT N On the bit lines BL0, BL N Generate on.
[0099] Fig.25 Depicted is a neuron VMM array 2500 that is particularly suitable for use with the memory cell 310 shown in FIG. 3 and serves as a synapse and component of neurons between an input layer and the next layer. In this example, inputs INPUT0, ..., INPUT M On word lines WL0, ..., WL M is received and output OUTPUT0, ..., OUTPUT N On the bit lines BL0, BL N Generate on.
[0100] Fig.26 Depicted is a neuron VMM array 2600 that is particularly suitable for use with the memory cell 410 shown in FIG. 4 and serves as a synapse and component of neurons between an input layer and the next layer. In this example, inputs INPUT0, ..., INPUT n On the vertical control gate lines CG0, CG NThe signals are received on the source lines SL0 and SL1, and outputs OUTPUT1 and OUTPUT2 are generated on the source lines SL0 and SL1.
[0101] Fig. 27 Depicted is a neuron VMM array 2700 that is particularly suitable for use with the memory cell 410 shown in FIG. 4 and serves as a synapse and component of neurons between an input layer and the next layer. In this example, inputs INPUT0, ..., INPUT N are received at the gates of bit line control gates 2701-1, 2701-2, ..., 2701-(N-1), and 2701-N, which are respectively coupled to bit lines BL0, ..., BL N Example outputs OUTPUT1 and OUTPUT2 are generated on source lines SL0 and SL1.
[0102] Fig.28 Depicted is a neuron VMM array 2800 that is particularly suitable for use with memory cells 310 shown in FIG. 3 , memory cells 510 shown in FIG. 5 , and Figure 7 The memory unit 710 shown is used as a synapse and component of the neurons between the input layer and the next layer. In this example, the inputs INPUT0, ..., INPUT M On word lines WL0, ..., WL M is received and output OUTPUT0, ..., OUTPUT N On bit lines BL0, BL N Generate on.
[0103] Fig.29 Depicted is a neuron VMM array 2900 that is particularly suitable for use with memory cells 310 shown in FIG. 3 , memory cells 510 shown in FIG. 5 , and Figure 7 The memory unit 710 shown is used as a synapse and component of the neurons between the input layer and the next layer. In this example, the inputs INPUT0, ..., INPUT M On the control gate lines CG0, CG M OUTPUT0, OUTPUT1, OUTPUT2, OUTPUT3, OUTPUT4 N On the vertical source lines SL0, . . . , SL N Each source line SL i The source line coupled to all memory cells in column i.
[0104] Fig.30 Depicted is a neuron VMM array 3000, which is particularly suitable for use with memory cells 310 shown in FIG. 3, memory cells 510 shown in FIG. 5, and Figure 7 The memory unit 710 shown is used as a synapse and component of the neurons between the input layer and the next layer. In this example, the inputs INPUT0, ..., INPUT M On the control gate lines CG0, CG M OUTPUT0, OUTPUT1, OUTPUT2, OUTPUT3, OUTPUT4 N On the vertical bit lines BL0, BL N On generation, each bit line BL i The bit line coupled to all memory cells in column i.
[0105] Long Short-Term Memory
[0106] Prior art includes a concept called long short-term memory (LSTM). LSTM cells are commonly used in neural networks. LSTM allows a neural network to remember information for a predetermined arbitrary time interval and use the information in subsequent operations. A conventional LSTM cell includes a cell, an input gate, an output gate, and a forget gate. The three gates regulate the flow of information into and out of the cell and the time interval at which information is remembered in the LSTM. VMMs are particularly useful in LSTM cells.
[0107] FIG. 14 depicts an example LSTM 1400. The LSTM 1400 in this example includes units 1401, 1402, 1403, and 1404. Unit 1401 receives an input vector x0 and generates an output vector h0 and a unit state vector c0. Unit 1402 receives an input vector x1, an output vector (hidden state) h0 from unit 1401, and a unit state c0 from unit 1401, and generates an output vector h1 and a unit state vector c1. Unit 1403 receives an input vector x2, an output vector (hidden state) h1 from unit 1402, and a unit state c1 from unit 1402, and generates an output vector h2 and a unit state vector c2. Unit 1404 receives an input vector x3, an output vector (hidden state) h2 from unit 1403, and a unit state c2 from unit 1403, and generates an output vector h3. Additional units may be used, and an LSTM with four units is merely an example.
[0108] Figure 15 depicts an example implementation of an LSTM cell 1500 that may be used for cells 1401, 1402, 1403, and 1404 in Figure 14. LSTM cell 1500 receives an input vector x(t), a cell state vector c(t-1) from a previous cell, and an output vector h(t-1) from a previous cell, and generates a cell state vector c(t) and an output vector h(t).
[0109] LSTM cell 1500 includes sigmoid function devices 1501, 1502, and 1503, each of which applies a number between 0 and 1 to control how much of each component in the input vector is allowed to pass through to the output vector. LSTM cell 1500 also includes tanh devices 1504 and 1505 for applying a hyperbolic tangent function to the input vector, multiplier devices 1506, 1507, and 1508 for multiplying two vectors together, and an addition device 1509 for adding two vectors together. The output vector h(t) can be provided to the next LSTM cell in the system, or it can be accessed for other purposes.
[0110] Fig.16 LSTM cell 1600 is depicted, which is an example of a specific implementation of LSTM cell 1500. For the convenience of the reader, the same numbering as LSTM cell 1500 is used in LSTM cell 1600. Sigmoid function devices 1501, 1502 and 1503 and tanh device 1504 each include multiple VMM arrays 1601 and activation function blocks 1602. Therefore, it can be seen that VMM arrays are particularly useful in LSTM cells used in certain neural network systems. Multiplier devices 1506, 1507 and 1508 and addition device 1509 are implemented digitally or in analog. Activation function block 1602 can be implemented digitally or in analog.
[0111] An alternative form of LSTM cell 1600 (and another example of a specific implementation of LSTM cell 1500) is Fig.17 In Fig.17 , the sigmoid function devices 1501, 1502 and 1503 and the tanh device 1504 share the same physical hardware (VMM array 1701 and activation function block 1702) in a time-division multiplexing manner. The LSTM unit 1700 also includes a multiplier device 1703 for multiplying two vectors together, an addition device 1708 for adding two vectors together, a tanh device 1505 (which includes an activation function block 1702), a register 1707 for storing the value i(t) when the value i(t) is output from the sigmoid function block 1702, a register 1704 for storing the value f(t)*c(t-1) when the value is output from the multiplier device 1703 through the multiplexer 1710, a register 1705 for storing the value i(t)*u(t) when the value is output from the multiplier device 1703 through the multiplexer 1710, and a register 1706 for storing the value o(t)*c~(t) when the value is output from the multiplier device 1703 through the multiplexer 1710, and a multiplexer 1709.
[0112] LSTM cell 1600 includes multiple sets of VMM arrays 1601 and corresponding activation function blocks 1602, while LSTM cell 1700 includes only one set of VMM arrays 1701 and activation function blocks 1702, which are used to represent multiple layers in the example of LSTM cell 1700. LSTM cell 1700 will require less space than LSTM 1600 because LSTM cell 1700 only requires 1 / 4 of its space for VMM and activation function blocks compared to LSTM cell 1600.
[0113] It is also understood that the LSTM cell will typically include multiple VMM arrays, each of which requires functionality provided by certain circuit blocks outside the VMM array (such as summers and activation function blocks and high voltage generation blocks). Providing a separate circuit block for each VMM array would require a large amount of space within the semiconductor device and would be inefficient to a certain extent. Therefore, the examples described below reduce the circuitry required outside the VMM array itself.
[0114] Gate Controlled Recursive Unit
[0115] The simulated VMM implementation can be used for a Gated Recurrent Unit (GRU) system. GRU is a gate mechanism in a recurrent neural network. GRU is similar to LSTM, except that a GRU cell generally contains fewer components than an LSTM cell.
[0116] FIG. 18 depicts an example GRU 1800. The GRU 1800 in this example includes units 1801, 1802, 1803, and 1804. Unit 1801 receives an input vector x0 and generates an output vector h0. Unit 1802 receives an input vector x1, an output vector h0 from unit 1801, and generates an output vector h1. Unit 1803 receives an input vector x2 and an output vector (hidden state) h1 from unit 1802 and generates an output vector h2. Unit 1804 receives an input vector x3 and an output vector (hidden state) h2 from unit 1803 and generates an output vector h3. Additional units may be used, and a GRU with four units is merely an example.
[0117] FIG. 19 depicts an example implementation of a GRU cell 1900 that may be used for cells 1801, 1802, 1803, and 1804 of FIG. 18. GRU cell 1900 receives an input vector x(t) and an output vector h(t-1) from a previous GRU cell, and generates an output vector h(t). GRU cell 1900 includes sigmoid function devices 1901 and 1902, each of which applies a number between 0 and 1 to components from the output vector h(t-1) and the input vector x(t). GRU cell 1900 also includes a tanh device 1903 for applying a hyperbolic tangent function to the input vector, a plurality of multiplier devices 1904, 1905, and 1906 for multiplying two vectors together, an addition device 1907 for adding two vectors together, and a complement device 1908 for subtracting the input from 1 to generate an output.
[0118] Fig. 20 A GRU unit 2000 is depicted, which is an example of a specific implementation of the GRU unit 1900. For the convenience of the reader, the same numbering is used in the GRU unit 2000 as in the GRU unit 1900. Fig. 20 As can be seen in FIG. 1 , sigmoid function devices 1901 and 1902 and tanh device 1903 each include multiple VMM arrays 2001 and activation function blocks 2002. Therefore, it can be seen that the VMM array is particularly useful in the GRU unit used in some neural network systems. Multiplier devices 1904, 1905 and 1906, addition device 1907 and complementary device 1908 are implemented in digital or analog form. Activation function block 2002 can be implemented in digital or analog form.
[0119] An alternative form of GRU unit 2000 (and another example of a specific implementation of GRU unit 1900) is Fig.21 In Fig.21 In , the GRU unit 2100 utilizes a VMM array 2101 and an activation function block 2102 which, when configured as a sigmoid function, applies a number between 0 and 1 to control how much of each component in the input vector is allowed to pass through to the output vector. Fig.21, the sigmoid function devices 1901 and 1902 and the tanh device 1903 share the same physical hardware (VMM array 2101 and activation function block 2102) in a time-division multiplexing manner. The GRU unit 2100 also includes a multiplier device 2103 for multiplying two vectors together, an addition device 2105 for adding two vectors together, a complementary device 2109 for subtracting an input from 1 to generate an output, a multiplexer 2104, a register 2106 for holding the value h(t-1)*r(t) when it is output from the multiplier device 2103 through the multiplexer 2104, a register 2107 for holding the value h(t-1)*z(t) when it is output from the multiplier device 2103 through the multiplexer 2104, and a register 2108 for holding the value h^(t)*(1-z(t)) when it is output from the multiplier device 2103 through the multiplexer 2104.
[0120] The GRU unit 2000 includes multiple sets of VMM arrays 2001 and activation function blocks 2002, while the GRU unit 2100 includes only one set of VMM arrays 2101 and activation function blocks 2102, which are used to represent multiple layers in the example of the GRU unit 2100. The GRU unit 2100 will require less space than the GRU unit 2000 because the GRU unit 2100 only requires 1 / 3 of its space for the VMM and activation function blocks compared to the GRU unit 2000.
[0121] It will also be appreciated that a GRU system will typically include multiple VMM arrays, each of which requires functionality provided by certain circuit blocks outside the VMM array (such as summers and activation function blocks and high voltage generation blocks). Providing a separate circuit block for each VMM array would require a significant amount of space within the semiconductor device and would be somewhat inefficient. Therefore, the examples described below reduce the circuitry required outside the VMM array itself.
[0122] The inputs to the VMM array can be analog levels, binary levels, pulses, time-modulated pulses, or digital bits (in which case a DAC is required to convert the digital bits to the appropriate input analog levels), and the outputs can be analog levels, binary levels, timed pulses, pulses, or digital bits (in which case an output ADC is required to convert the output analog levels to digital bits).
[0123] In general, for each memory cell in the VMM array, each weight W can be implemented by a single memory cell or by a differential cell or by two hybrid memory cells (average of 2 cells). In the case of a differential cell, two memory cells are required to implement the weight W as a differential weight (W=W+-W-). In two hybrid memory cells, two memory cells are required to implement the weight W as the average of the two cells.
[0124] Fig.31 A VMM system 3100 is depicted. In some examples, the weights W stored in the VMM array are stored as a differential pair W+ (positive weight) and W- (negative weight), where W=(W+)-(W-). In the VMM system 3100, half of the bit lines are designated as W+ lines, i.e., bit lines connected to memory cells that will store positive weights W+, and the other half of the bit lines are designated as W- lines, i.e., bit lines connected to memory cells that implement negative weights W-. The W- lines are interspersed between the W+ lines in an alternating manner. The subtraction operation is performed by a summing circuit (such as summing circuits 3101 and 3102), which receives current from the W+ line and the W- line. The output of the W+ line and the output of the W- are combined together to effectively give W=W+-W- for each (W+, W-) cell pair of all (W+, W-) line pairs. Although described above with respect to W- lines interspersed between W+ lines in an alternating manner, in other examples, the W+ lines and W- lines may be arbitrarily located anywhere in the array.
[0125] Fig.32 Another example is depicted. In VMM system 3210 , positive weights W+ are implemented in a first array 3211 , and negative weights W− are implemented in a second array 3212 , which is separate from the first array, and the resulting weights are appropriately combined together by summing circuit 3213 .
[0126] Fig.33A VMM system 3300 is depicted, in which the weights W stored in the VMM array are stored as a differential pair W+ (positive weight) and W- (negative weight), where W=(W+)-(W-). The VMM system 3300 includes an array 3301 and an array 3302. Half of the bit lines in each of the arrays 3301 and 3302 are designated as W+ lines, i.e., bit lines connected to memory cells that will store positive weights W+, and the other half of the bit lines in each of the arrays 3301 and 3302 are designated as W- lines, i.e., bit lines connected to memory cells that implement negative weights W-. The W- lines are interspersed between the W+ lines in an alternating manner. The subtraction operation is performed by a summing circuit (such as summing circuits 3303, 3304, 3305, and 3306), which receives current from the W+ lines and the W- lines. The outputs of the W+ lines and the outputs of the W- lines from each array 3301, 3302 are combined together to effectively give W=W+-W- for each (W+, W-) cell pair for all (W+, W-) line pairs. In addition, the W values from each array 3301 and 3302 can be further combined by summing circuits 3307 and 3308 so that each W value is the result of subtracting the W value from array 3301 from the W value from array 3302, which means that the final result from summing circuits 3307 and 3308 is a difference of two difference values.
[0127] Each nonvolatile memory cell used in an analog neural memory system is erased and programmed to hold a very specific and precise amount of charge (i.e., number of electrons) in the floating gate. For example, each floating gate should hold one of N different values, where N is the number of different weights that can be indicated by each cell. Examples of N include 16, 32, 64, 128, and 256.
[0128] Fig.34 A block diagram of a VMM system 3400 is depicted. The VMM system 3400 includes a VMM array 3401, a row decoder 3402, a high voltage decoder 3403, a column decoder 3404, a bit line driver 3405, an input circuit 3406, an output circuit 3407, a control logic unit 3408, and a bias generator 3409. The VMM system 3400 further includes a high voltage generation block 3410, which includes a charge pump 3411, a charge pump regulator 3412, and a high voltage analog precision level generator 3413. The VMM system 3400 further includes a (program / erase, or weight tuning) algorithm controller 3414, an analog circuit 3415, a control engine 3416 (which may include special functions such as arithmetic functions, activation functions, embedded microcontroller logic, etc., without limitation) and a test control logic unit 3417.
[0129] The input circuit 3406 may include various circuits such as a DAC (digital-to-analog converter), a DPC (digital-to-pulse converter, digital-to-time modulated pulse converter), an AAC (analog-to-analog converter, such as a current-to-voltage converter, a logarithmic converter), a PAC (pulse-to-analog level converter), or any other type of converter. The input circuit 3406 may implement one or more of normalization, linear or nonlinear up / down scaling functions, or arithmetic functions. The input circuit 3406 may implement a temperature compensation function for the input level. The input circuit 3406 may implement an activation function such as a ReLU or a sigmoid.
[0130] Output circuit 3407 may include circuits such as an ADC (analog-to-digital converter for converting neuron analog outputs into digital bits), an AAC (analog-to-analog converter, such as a current-to-voltage converter, a logarithmic converter), an APC (analog-to-pulse converter, analog-to-time modulated pulse converter), or any other type of converter. Output circuit 3407 may implement an activation function, such as a rectified linear activation function (ReLU) or a sigmoid. Output circuit 3407 may implement one or more of statistical normalization, regularization, up / down scaling / gain functions, statistical rounding, or arithmetic functions (e.g., addition, subtraction, division, multiplication, shift, logarithm) for the neuron output. Output circuit 3407 may implement a temperature compensation function for the neuron output or array output (such as a bit line output) to keep the power consumption of the array approximately constant, or to improve the accuracy of the array (neuron) output, such as by keeping the IV slope approximately the same.
[0131] As the application of artificial neural networks becomes more complex, larger VMM arrays are increasingly needed. At the same time, there is a need to efficiently use the space within the packaged integrated circuit and save as much power as possible while still maintaining accuracy so that each of the N different weights is still correctly stored and read. Summary of the invention
[0132] Several examples are described for providing an artificial neural network system including a three-dimensional integrated circuit that includes one or more VMM arrays. BRIEF DESCRIPTION OF THE DRAWINGS
[0133] FIG. 1 is a diagram illustrating an artificial neural network.
[0134] FIG. 2 depicts a prior art split-gate flash memory cell.
[0135] FIG. 3 depicts another prior art split gate flash memory cell.
[0136] FIG. 4 depicts another prior art split gate flash memory cell.
[0137] FIG. 5 depicts another prior art split gate flash memory cell.
[0138] Figure 6 A diagram illustrating different levels of an example artificial neural network utilizing one or more non-volatile memory arrays.
[0139] Figure 7 FIG. 1 is a block diagram showing a VMM system.
[0140] Figure 8 is a block diagram illustrating an example artificial neural network utilizing one or more VMM systems.
[0141] Fig. 9 Another example of a VMM system is depicted.
[0142] Fig.10 Another example of a VMM system is depicted.
[0143] Fig.11 Another example of a VMM system is depicted.
[0144] Fig.12 Another example of a VMM system is depicted.
[0145] Fig.13 Another example of a VMM system is depicted.
[0146] FIG. 14 depicts a prior art long short term memory system.
[0147] FIG. 15 depicts an example cell for use in a long short term memory system.
[0148] Fig.16 An example implementation of the unit of FIG. 15 is depicted.
[0149] Fig.17 Another example implementation of the unit of FIG. 15 is depicted.
[0150] FIG. 18 depicts a prior art gate-controlled recursive cell system.
[0151] FIG. 19 depicts an example cell for use in a gate-controlled recursive cell system.
[0152] Fig. 20 An example implementation of the unit of FIG. 19 is depicted.
[0153] Fig.21 Another example implementation of the unit of FIG. 19 is depicted.
[0154] Fig. 22 Another example of a VMM system is depicted.
[0155] Fig.23Another example of a VMM system is depicted.
[0156] Fig.24 Another example of a VMM system is depicted.
[0157] Fig.25 Another example of a VMM system is depicted.
[0158] Fig.26 Another example of a VMM system is depicted.
[0159] Fig. 27 Another example of a VMM system is depicted.
[0160] Fig.28 Another example of a VMM system is depicted.
[0161] Fig.29 Another example of a VMM system is depicted.
[0162] Fig.30 Another example of a VMM system is depicted.
[0163] Fig.31 Another example of a VMM system is depicted.
[0164] Fig.32 Another example of a VMM system is depicted.
[0165] Fig.33 Another example of a VMM system is depicted.
[0166] Fig.34 An example of a 2D VMM system is depicted.
[0167] Fig.35 An example of a 3D VMM system is depicted.
[0168] Fig.36 An example of a 3D VMM system is depicted.
[0169] Fig.37A and Fig.37B An example of a 3D VMM system is depicted.
[0170] Fig.38 An example of a 3D VMM system is depicted.
[0171] Fig.39A , Fig.39B and Fig.39C An example of a 3D VMM system is depicted.
[0172] Fig.40 An example of a 3D VMM system is depicted.
[0173] Fig.41An example of a 3D VMM system is depicted.
[0174] Fig.42 An example of a 3D VMM system is depicted.
[0175] Fig.43 An example of a 3D VMM system is depicted.
[0176] Fig.44 An example of a 3D VMM system is depicted.
[0177] Fig.45 An example of a 3D VMM system is depicted.
[0178] Fig.46 An example of a 3D VMM system is depicted.
[0179] Fig.47 An example of a 3D VMM system is depicted.
[0180] Fig.48 An example of a 3D VMM system is depicted.
[0181] Fig.49 An example of a neuronal circuit is depicted.
[0182] Fig.50 An example of a neuronal circuit is depicted.
[0183] Fig.51 An example of a drive circuit is depicted.
[0184] Fig.52 An example of a differential neuron circuit is depicted.
[0185] Fig.53 An example of a sample-and-hold buffer is depicted.
[0186] Fig.54 An example of an ADC circuit is depicted.
[0187] Fig.55 An example of an ADC circuit is depicted. DETAILED DESCRIPTION
[0188] 3D VMM System Architecture
[0189] Fig.35 Depicted is a 3D VMM system 3500 that includes multiple dies, such as dies 3501, 3502, 3503, 3504, 3505, and 3506, that are vertically stacked within a package 3522 to form a packaged integrated circuit. The 3D VMM system 3500 includes certain functional blocks that are functionally similar to those contained in the Fig.343501 and 3502 share components contained in dies 3503, 3504, 3505, and 3506.
[0190] In this example, die 3501 includes a corresponding VMM array 3507 (similar in function to Fig.34 VMM 3401 in FIG. 1 ), a corresponding input multiplexer 3509, a corresponding row buffer 3523 (which may, for example, provide a sample-and-hold buffer voltage to the array input), a corresponding high voltage decoder 3508 (similar in function to Fig.34 3D VMM system 3500) and corresponding neuron circuits 3510 (which may, for example, perform scaling functions, minimum / maximum limit functions, differential output conversion, buffering of array output current, but are not limited thereto). Input multiplexer 3509 receives analog input signals and applies them to VMM array 3507, and neuron circuits 3510 receive analog output signals representing neuron outputs from VMM array 3507. Output signals may be sent to other blocks within 3D VMM system 3500.
[0191] The die 3502 also includes a corresponding VMM array 3507 , a corresponding input multiplexer 3509 , a corresponding row buffer 3522 , a corresponding high voltage decoder 3508 , and a corresponding neuron circuit 3510 .
[0192] In this example, two dies (die 3501 and 3502) contain corresponding VMM arrays 3507, but it should be understood that additional dies containing corresponding VMM arrays may be included.
[0193] Die 3503 includes a high voltage generator 3511 (similar in function to Fig.34 The high voltage generation block 3410 in FIG. 1 and the analog circuit 3512 (similar in function to Fig.343D VMM system 3500 may have thermal challenges that 2D VMM system 3400 does not have. Since die 3501, 3502, 3503, 3504, 3505, and 3506 are stacked in a vertical configuration and contain different types of circuits, each die may experience different thermal operating conditions during operation. For example, a die will become hotter than other die, and the rate of temperature increase may vary between die. This introduces the possibility of inaccuracies due to thermal variations. Temperature compensation circuit 3513 compensates for temperature variations experienced between individual die. Optionally, one or more thermal sensors are located on each respective die to provide temperature data to temperature compensation circuit 3513. Temperature compensation circuit 3513 then changes fine tuning or configuration settings to compensate for any changes in temperature. Temperature compensation circuit 3513 also compensates for temperature variations of each die, for example to compensate for changes in cell current with temperature, such as making the resulting bit line (neuron) current approximately the same with temperature. Temperature compensation circuit 3513 is also used for neuron circuit 3510, DAC and ADC circuits to make the operating dynamic range (e.g., output range of DAC, input range of ADC, output range of neuron circuit 3510) approximately the same over temperature.
[0194] Die 3504 contains input circuitry 3514 (functionally similar to input circuitry 3406) which includes address decoding circuitry 3524, row registers 3525 (holding active input values for array rows), and digital-to-analog converters (DACs) 3515. DACs 3515 receive digital signals from row registers 3525 and convert them to analog signals.
[0195] Die 3505 includes an analog-to-digital converter (ADC) 3516. ADC 3516 receives an analog signal and converts it into a digital signal.
[0196] The die 3506 includes a digital circuit 3517, a static random access memory (SRAM) 3518, a register 3519, a physical I / O connection 3520, a digital accelerator 3531, and an on-chip network (NOC) 3715. The die 3506 provides control functions for other dies. The digital circuit 3517 may include digital logic, a microcontroller, a SIMD (single instruction multiple data) processor, and a processor. The SRAM 3518 and the register 3519 may be used to store system information and configuration information used by the digital circuit 3517 or other circuits or blocks in the 3D VMM system 3500. The physical I / O connection 3520 provides an IO interface to a device outside the VMM system 3500 (such as an external processing unit) or another package 3522. The digital accelerator 3531 is used for certain neural networks or certain layers within a neural network that may require additional processing, such as when there is a small activation size, where the weights stored in the unit are dynamic rather than fixed, where a MAC operation needs to be performed without restrictions. . NOC 3715 provides network routing functionality within 3D VMM system 3500 , for example, by generating control signals to route signals from one block to another.
[0197] A corresponding die in the plurality of dies is connected to one or more other dies in the plurality of dies through a vertical interface 3521, which respectively connects two or more dies together. In one example, the vertical interface 3521 is implemented as a through silicon via (TSV).
[0198] During a read operation of the 3D VMM system 3500, a digital input is received by the input circuit 3514. The digital input enables the row register 3525, which stores the activation input and applies the selected activation input to the DAC 3515 in response to the digital input, which converts the digital output from the row register 3525 into a corresponding analog signal. The analog signal generated by the DAC 3515 is provided by the input circuit 3514 through one or more vertical interfaces 3521 to the input multiplexer 3509 and row buffer 3523 on one or more of the dies 3501, 3502, which then applies the signal to one or more rows in the corresponding VMM array 3507, resulting in the generation of output by the VMM array 3507. The output from the corresponding VMM array 3507 is received by the corresponding neuron circuit 3510, which provides a buffering function to drive the parasitic capacitance of the one or more vertical interfaces 3521 to which it is connected. The neuron circuit 3510 provides an analog signal to the ADC 3516 on the die 3505 through one or more vertical interfaces 3521, which converts the analog signal into a digital signal. Alternatively, the analog signal can bypass the ADC 3516 and remain in analog form. The output of the ADC 3516 is provided to a device (such as a processing unit or a graphics processing unit) outside the 3D VMM system 3500 through a physical I / O 3520, or is applied as an input to a corresponding VMM array 3507 (representing another layer in the artificial neural network). Alternatively, the analog signal from the neuron circuit 3510 can bypass the ADC 3516 and remain in analog form and be applied as an input to the corresponding VMM array 3507.
[0199] Fig.36 3D VMM system 3600 is depicted including package 3622. 3D VMM system 3600 is similar to 3D VMM system 3500 and contains many of the same components, except for some differences in the arrangement of components and certain additional components. Fig.35 The same project contains Fig.36 Here, components included in dies 3601 and 3602 share components included in dies 3603, 3604, 3605, and 3606.
[0200] 3D VMM system 3600 includes multiple dies, such as dies 3601 , 3602 , 3603 , 3604 , 3605 , and 3606 , which are vertically stacked within a common package 3522 to form a packaged integrated circuit.
[0201] In this example, die 3601 includes a corresponding VMM array 3507 , a corresponding input multiplexer 3509 , a corresponding register 3524 (holding the active input value of the array row), a corresponding row buffer 3523 , a corresponding high voltage multiplexer 3608 , and a corresponding column multiplexer 3610 .
[0202] The die 3602 also includes a corresponding VMM array 3507 , a corresponding input multiplexer 3509 , a corresponding register 3524 , a corresponding row buffer 3523 , a corresponding high voltage multiplexer 3608 , and a column multiplexer 3610 .
[0203] In this example, two dies (die 3601 and 3602) contain respective VMM arrays 3507, but it should be understood that additional dies containing VMM arrays may be included.
[0204] The die 3603 includes a high voltage generator 3511 , an analog circuit 3512 , a temperature compensation circuit 3513 , and a high voltage multiplexer 3608 .
[0205] Die 3604 includes input circuitry 3614 , which includes address decode 3524 and DAC 3515 .
[0206] Die 3605 includes neuron circuit 3510 and ADC 3516.
[0207] The bare die 3606 includes digital circuits 3517 , SRAM 3518 , registers 3519 , and physical I / O connections 3520 .
[0208] A respective die in the plurality of dies is connected to one or more other dies within the plurality of dies through a respective vertical interface 3521 .
[0209] During a read operation of the 3D VMM system 3600, input is received by the input circuit 3614 and an address is received. The address decoder 3524 decodes the address and provides input to a corresponding row register 3525 corresponding to the decoded address through one or more vertical interfaces 3521. The output of the corresponding row register 3525 is coupled to the DAC 3515 through one or more vertical interfaces 3521, which converts the digital output bits received from the row register 3525 into an analog signal and provides the analog signal to the corresponding input multiplexer 3509 and the row buffer 3523 through one or more vertical interfaces 3521. The row buffer 3523 applies the buffered analog signal to the row input (such as an array control gate or word line) of the corresponding VMM array 3507 of the selected row. The output from the corresponding VMM array 3507 (such as from the array bit line) is received by the corresponding column multiplexer 3610, and the output of the corresponding column multiplexer 3610 provides a signal to the ADC 3516 on the die 3605 through one or more vertical interfaces 3521, which converts the analog signal into a digital signal. Alternatively, the signal can bypass the ADC 3516 and remain in analog form. The output of the ADC 3516 is provided to the digital circuit 3517 on the die 3606 (which can perform an activation function, a pooling function, or other network functions) through one or more vertical interfaces 3521, and the output of the digital circuit 3517 can be provided to a device (such as a processing unit or a graphics processing unit) external to the 3D VMM system 3600 through the physical I / O 3520 on the die 3606, or provided to the input circuit 3614 on the die 3604 through the corresponding vertical interface 3521 as an input to another corresponding VMM array 3507 (representing another layer in the artificial neural network).
[0210] Fig.37AA 3D VMM system 3700 is depicted that includes multiple dies in two or more vertical stacks. In this example, the first vertical stack includes dies 3701, 3702, 3703, 3704, 3705, and 3706, and the second vertical stack includes dies 3707, 3708, 3709, 3710, 3711, and 3712, all of which are contained in a common package 3522 to form a single packaged integrated circuit. In one example, the dies in the first vertical stack are physically separated from the dies in the second vertical stack. In another example, the dies in the first vertical stack and the dies in the second vertical stack are the same physical dies (meaning, for example, that dies 3701 and 3707 are the same die). Here, components included in dies 3701 and 3702 share components included in dies 3703 , 3704 , 3705 , and 3706 , and components included in dies 3707 and 3708 share components included in dies 3709 , 3710 , 3711 , and 3712 .
[0211] In the example shown, die 3701 and die 3707 each include a corresponding VMM array 3507, a corresponding array input 3729 (which includes a corresponding input multiplexer 3509, a corresponding address decoder 3713, a corresponding row register 3525 and a corresponding row buffer 3523), a corresponding high voltage multiplexer 3608 and a corresponding neuron circuit 3510.
[0212] Die 3702 and die 3708 also each include a corresponding VMM array 3507 , a corresponding array input 3729 (which includes a corresponding input multiplexer 3509 , a corresponding address decoder 3713 , a corresponding row register 3525 , and a corresponding row buffer 3523 ), a corresponding high voltage multiplexer 3608 , and a corresponding neuron circuit 3510 .
[0213] In this example, four dies (die 3701, 3702, 3707, and 3708) contain respective VMM arrays 3507, but it is understood that additional dies containing VMM arrays may be included.
[0214] Die 3703 and die 3709 each include a corresponding high voltage decoder 3714 , a corresponding high voltage generator 3511 , a corresponding analog circuit 3512 , and a corresponding temperature compensation circuit 3513 .
[0215] Die 3704 and die 3710 each include a respective input circuit 3514 , which includes a respective DAC 3515 .
[0216] Die 3705 and die 3711 each include a corresponding ADC 3516 .
[0217] Die 3706 and die 3712 each include respective digital circuits 3517 , respective SRAM 3518 , respective registers 3519 , respective physical I / O connections 3520 , and respective NOC (network on chip) connections 3715 .
[0218] A corresponding die in the plurality of dies is connected to one or more other dies in the plurality of dies through one or more vertical interfaces 3521 or horizontal interfaces 3716, which respectively connect two or more dies together. In one example, the vertical interface 3521 is a through silicon via (TSV). In one example, the horizontal interface 3716 is a redistribution layer (RDL) connection portion.
[0219] During a read operation of the 3D VMM system 3700, a digital input is received by a corresponding input circuit 3514, which converts the digital input into an analog signal via its corresponding DAC 3515, and the output of the corresponding DAC 3515 is coupled to the row register 3525 of the corresponding array input 3729 via the corresponding vertical interface 3521 and / or the horizontal interface 3716. The output of the row register 3525 of the corresponding array input 3729 is provided to the input multiplexer 3509 and the row buffer 3523, which then applies the signal to one or more rows in the VMM array 3507. The output from the corresponding VMM array 3507 is received by the corresponding neuron circuit 3510, which provides a buffering function to drive the parasitic capacitance of the one or more vertical interfaces 3521 or horizontal interfaces 3716 to which it is connected. The neuron circuit 3510 provides a buffered analog signal to a corresponding ADC 3516 through one or more vertical interfaces 3521 or horizontal interfaces 3716, and the ADC converts the analog signal into a digital signal. The output of the ADC 3516 is provided to a corresponding digital circuit 3517 (which performs an activation function, a pooling function, or a network function), and the output of the corresponding digital circuit 3517 can be provided to a device (such as a processing unit or a graphics processing unit) outside the 3DVMM system 3700 through a corresponding physical I / O 3520 or provided to a corresponding input circuit 3514 to be converted by a corresponding DAC 3515, and the output of the corresponding input circuit 3514 is coupled to another VMM array 3507 (representing another layer in the artificial neural network) or coupled to another package 3522 through a corresponding physical I / O 3520.
[0220] Fig.37BDepicted is a 3D VMM system 3750 that is similar to 3D VMM system 3750 except that it has another type of VMM array shown on die 3758 as VMM array 3557 that includes either static RAM cells or dynamic RAM cells.
[0221] Fig.38 Depicted is a 3D VMM system 3800, which includes multiple dies in two vertical stacks. There can be more than two stacks. In this example, the first vertical stack includes dies 3801, 3802, 3803, 3804, 3805 and 3806, and the second vertical stack includes dies 3807, 3808, 3809, 3810, 3811 and 3812, all of which are contained in a common package 3522 to form a single packaged integrated circuit. In one example, the die in the first vertical stack is a die physically separated from the die in the second vertical stack. In another example, the die in the first vertical stack and the die in the second vertical stack are the same physical die (meaning, for example, that die 3801 and 3807 are the same die). Here, components included in dies 3801 and 3802 share components included in dies 3803 , 3804 , 3805 , and 3806 , and components included in dies 3807 and 3808 share components included in dies 3809 , 3810 , 3811 , and 3812 .
[0222] In the example shown, die 3801, 3802, 3807, and die 3808 each include a respective VMM array 3507, a respective array input 3729, a respective high voltage multiplexer 3608, a respective column multiplexer 3610, and a respective array input circuit 3729. The respective array input 3729 includes an input multiplexer 3509, an address decoder 3713, a row register 3524, and a row buffer 3523.
[0223] In this example, four dies (die 3801, 3802, 3807, and 3808) contain VMM arrays, but it should be understood that additional dies containing VMM arrays may be included.
[0224] The die 3803 and the die 3809 include a high voltage decoder 3714, a high voltage generator 3511, an analog circuit 3512, and a temperature compensation circuit 3513, respectively.
[0225] Die 3804 and die 3810 each include input circuit 3514 including DAC 3515 .
[0226] Die 3805 and die 3811 include ADC 3516 and neuron circuit 3510, respectively.
[0227] Die 3806 and die 3812 include digital circuits 3517, SRAM 3518, registers 3519, physical I / O connections 3520, and NOC connections 3715, respectively.
[0228] Multiple dies are connected to one or more other dies within the multiple dies through one or more vertical interfaces 3521 or horizontal interfaces 3716, respectively, and the one or more vertical interfaces or horizontal interfaces respectively connect two or more dies together. In one example, the corresponding vertical interface 3521 is implemented as a through silicon via (TSV). In one example, the corresponding horizontal interface 3716 is implemented as a redistribution layer (RDL) connection.
[0229] During a read operation of the 3D VMM system 3800, a digital input is received by an input circuit 3514, which uses a DAC 3515 to convert the digital input into analog form and provides an analog signal to a corresponding array input circuit 3729 via a corresponding vertical interface 3521 and / or a horizontal interface 3716. The array input circuit 3729 receives the analog signal and the address from the input circuit, and then applies the analog signal to a selected row in the VMM array 3507 in response to the address. The output from the VMM array 3507 is received by a column multiplexer 3610, which provides an analog signal to an ADC 3516 through one or more vertical interfaces 3521 or horizontal interfaces 3716, which converts the analog signal into a digital signal. Alternatively, the analog signal may bypass the ADC 3516 and remain in analog form. The output of ADC 3516 is provided to digital circuit 3517 (which can execute activation function, pooling function or other network function) through one or more vertical interfaces 3521 or horizontal interface 3716, and the output of digital circuit 3517 can be provided to a device outside 3D VMM system 3800 (such as a processing unit or graphics processing unit) through physical I / O 3520, or provided to input circuit 3514 of other VMM array 3507 (representing another layer in artificial neural network), or provided to another package 3522 through physical I / O 3520.
[0230] Fig.39AA 3D VMM system 3900 is depicted, which includes multiple dies in two vertical stacks. In this example, the first vertical stack includes dies 3901, 3902, 3903, and 3904, and the second vertical stack includes dies 3905, 3906, 3907, and 3908, all of which are contained in a common package 3522 to form a single packaged integrated circuit. In one example, the die in the first vertical stack is a die physically separated from the die in the second vertical stack. In another example, the die in the first vertical stack and the die in the second vertical stack are the same physical die (meaning, for example, that die 3901 and 3905 are the same die). Here, the components contained in die 3901 and 3902 share the components contained in die 3903 and 3904, and the components contained in die 3905 and 3906 share the components contained in die 3907 and 3908.
[0231] In the example shown, die 3901, 3902, 3905, and die 3906 include VMM array 3507, array input 3729, high voltage multiplexer 3608, and column multiplexer 3610, respectively.
[0232] In this example, four dies (die 3901, 3902, 3903, and 3904) contain VMM arrays, but it should be understood that additional dies containing VMM arrays may be included.
[0233] Die 3903 and die 3907 include a high voltage decoder 3714, a high voltage generator 3511, an analog circuit 3512, a temperature compensation circuit 3513, an input circuit 3514 including a DAC 3515, a neuron circuit 3510, and an ADC 3516, respectively.
[0234] Die 3904 and die 3908 include digital circuits 3517, SRAM 3518, registers 3519, physical I / O connections 3520, and NOC connections 3715, respectively.
[0235] The plurality of dies are respectively connected to one or more other dies within the plurality of dies through one or more vertical interfaces 3521 or horizontal interfaces 3716 , which respectively connect two or more dies together.
[0236] During a read operation of the 3D VMM system 3900, a digital input is received by an input circuit 3514, which uses a DAC 3515 to convert the digital input into analog form and provides an analog signal to a corresponding array input circuit 3729 via a corresponding vertical interface 3521 and / or a horizontal interface 3716. The array input circuit 3729 receives the analog signal and an address from the input circuit, and applies the analog signal to a selected row in the VMM array 3507 in response to the received address. The output from the VMM array 3507 is received by a column multiplexer 3610, which provides an analog signal to an ADC 3516 through one or more vertical interfaces 3521 or horizontal interfaces 3716, which converts the analog signal into a digital signal. Alternatively, the analog signal may bypass the ADC 3516 and remain in analog form. The output of ADC 3516 is provided to digital circuit 3517 via corresponding vertical interface 3521 and / or horizontal interface 3716, and then provided to a device external to the 3D VMM system 3900 (such as a processing unit or a graphics processing unit) through physical I / O 3520 or provided to input circuit 3514 to be applied as input to VMM array 3507 (representing another layer in the artificial neural network) or provided to another package 3522 through physical I / O 3520.
[0237] Fig.39B Describes something like Fig.39A 3D VMM system 3950 of the present invention is shown in FIG. 3D VMM system 3950, except that dies 3951, 3952, 3955, and 3956 now have both input circuits 3514 and array inputs 3729. 3D VMM system 3950 includes package 3522 and dies 3951, 3952, 3953, 3954, 3955, 3956, 3957, and 3958. The plurality of dies are connected to one or more other dies within the plurality of dies through one or more vertical interfaces 3521 or horizontal interfaces 3716, respectively, which connect two or more dies together.
[0238] Fig.39CA 3D VMM system 3980 is depicted, which includes multiple dies in two vertical stacks. In this example, the first vertical stack includes dies 3981, 3982, 3983, and the second vertical stack includes dies 3984, 3985, and 3986, all of which are contained in a common package 3522 to form a single packaged integrated circuit. In one example, the die in the first vertical stack is a die physically separated from the die in the second vertical stack. In another example, the die in the first vertical stack and the die in the second vertical stack are the same physical die (meaning, for example, that die 3981 and 3984 are the same die). Here, the components contained in die 3981 and 3982 share the components contained in die 3983, and the components contained in die 3984 and 3985 share the components contained in die 3986.
[0239] In the example shown, die 3981, 3982, 3984, and die 3985 include a VMM array 3507, a high voltage block 3991, an input block 3990, an output block 3992, and an analog block 3993, respectively. Input block 3990 may include input circuit 3514, DAC 3515, array input circuit 3729. High voltage block 3991 may include a high voltage multiplexer 3608 and a high voltage decoder 3714. Output block 3992 may include a column multiplexer 3610, a neuron circuit 3510, and an ADC 3516. Analog block 3993 may include a high voltage generator 3511, an analog circuit 3512, and a temperature compensation circuit 3513.
[0240] Die 3983 and die 3986 each include digital circuits 3517 , SRAM 3518 , registers 3519 , physical I / O connections 3520 , digital accelerators 3521 (used digitally for multiple and accumulate (MAC) functions), and NOC connections 3715 .
[0241] The plurality of dies are respectively connected to one or more other dies within the plurality of dies through one or more vertical interfaces 3521 or horizontal interfaces 3716, each of which respectively connects two or more dies together.
[0242] Fig.40Depicted is a 3D VMM system 4000, which includes multiple dies in two vertical stacks. In this example, the first vertical stack includes dies 4001, 4002, 4003 and 4004, and the second vertical stack includes dies 4005, 4006, 4007 and 4008, all of which are contained in a common package 3522 to form a single package integrated circuit. In one example, the die in the first vertical stack is a die physically separated from the die in the second vertical stack. In another example, the die in the first vertical stack and the die in the second vertical stack are the same physical die (meaning, for example, that die 4001 and 4005 are the same die). Here, the components contained in die 4001 and 4002 share the components contained in die 4003 and 4004, and the components contained in die 4005 and 4006 share the components contained in die 4007 and 4008.
[0243] In the example shown, dies 4001 , 4002 , 4005 , and 4006 include VMM array 3507 , array input 4029 (which includes input multiplexer 3509 and / or decoder 3713 —not shown), high voltage multiplexer 3608 , and column multiplexer 3610 , respectively.
[0244] In this example, four dies (die 4001, 4002, 4003, and 4004) contain VMM arrays, but it should be understood that additional dies containing VMM arrays may be included.
[0245] Die 4003 and die 4007 include a high voltage decoder 3714, a high voltage generator 3511, an analog circuit 3512, a temperature compensation circuit 3513 and a neuron circuit 3510, respectively.
[0246] Die 4004 and die 4008 include digital circuits 3517, SRAM 3518, registers 3519, physical I / O connections 3520, and NOC connections 3715, respectively.
[0247] Unlike VMM system 3900, VMM system 4000 does not include DAC 3515 and ADC 3516 because the inputs and outputs remain in analog form and are not converted between analog and digital forms.
[0248] The plurality of dies are respectively connected to one or more other dies within the plurality of dies through one or more vertical interfaces 3521 or horizontal interfaces 3716 , which respectively connect two or more dies together.
[0249] During a read operation of the 3D VMM system 4000, an analog input (such as a voltage, current, or a timing-based entity such as a pulse train) is received by a corresponding array input 4029, which then applies the signal to one or more rows in the corresponding VMM array 3507. The output from the corresponding VMM array 3507 is received by a column multiplexer 3610, which provides the analog signal (such as a voltage, current, or a timing-based entity) to a corresponding neuron circuit 3510 through one or more vertical interfaces 3521 or horizontal interface 3716, and the neuron circuit 3510 provides a buffered signal through a physical I / O 3520 to a device external to the 3D VMM system 4000 (such as a processing unit or graphics processing unit) or another package 3522, or to the array input 4029 of another VMM array 3507 (representing another layer in the artificial neural network).
[0250] Figure 41 to Figure 44 Additional details regarding an example configuration of VMM array 3507 are depicted. Figure 41 to Figure 44 3D VMM systems 4100, 4200, 4300, and 4400 are depicted, respectively. Each of the 3DD VMM systems 4100, 4200, 4300, and 4400 includes a VMM array 3507-1 on a first die and a VMM array 3507-2 on a second die in a common package (not shown). VMM arrays 3507-1 and 3507-2 include arrays of non-volatile memory cells arranged in m+1 rows and n+1 columns, respectively. The corresponding rows are coupled to one of the control gate lines labeled CG0, ..., CGm, and the corresponding columns are coupled to one of the bit lines labeled BL0, ..., BLn. The cell is located at the intersection of the bit line and the control gate line. For example, cell 4101mn is located in row m and column n and is coupled to CGm and BLn in VMM array 3507-1, and cell 4102mn is located in row m and column n and is coupled to CGm and BLn in VMM array 3507-2. Since VMM arrays 3507-1 and 3507-2 are located on different dies, they can optionally be manufactured using different semiconductor processes. Regardless of whether the same or different semiconductor processes are used to manufacture VMM arrays 3507-1 and 3507-2, the cells in VMM array 3507-1 can store a different number of bits than the cells in VMM array 3507-2. For example, a cell in VMM array 3507-1 (such as cell 4101mn) can store i bits, while a cell in VMM array 3507-2 (such as cell 4102mn) can store j bits, where i and j are integers of different values. For example, i may be 3 (meaning that the cells in VMM array 3507-1 store 3-bit values respectively), and j may be 5 (meaning that the cells in VMM array 3507-2 store 5-bit values respectively).
[0251] exist Fig.41 In the embodiment, inputs are respectively provided to the VMM arrays 3507-1 and 3507-2 through control gate lines, and outputs are respectively obtained on bit lines. Fig.41 Example units 4101mn and 4102mn are depicted. Optionally, if desired, the outputs may be combined elsewhere, such as by converting them to digital form using ADC 3516 (as shown in previous figures) and adding them together using digital circuit 3517 (as shown in previous figures).
[0252] exist Fig.42 3507-2, respectively. The output of the VMM array 3507-1 is obtained on a first set of bit lines, and the output of the VMM array 3507-2 is obtained on a second set of bit lines, where the first set of bit lines are coupled to the second set of bit lines through corresponding vertical interfaces 3521, which effectively adds the output signals together in analog form. Optionally, if desired, the combined analog outputs can be digitized elsewhere, such as by converting them to digital form using ADC 3516 (as shown in previous figures).
[0253] exist Fig.43 , inputs are provided to VMM array 3507-1 through a first set of control gate lines, and inputs are provided to VMM array 3507-2 through a second set of control gate lines, wherein the first set of control gate lines are coupled to the second set of control gate lines through corresponding vertical interfaces 3521, and outputs are obtained on the bit lines, respectively. Optionally, if desired, the outputs can be combined elsewhere, such as by converting them to digital form using ADC 3516 (as shown in the previous figures), and adding them together using digital circuit 3517 (as shown in the previous figures).
[0254] exist Fig.44 3507-1 and 3507-2 are provided inputs together through control gate lines, through vertical interface 3521. Outputs are obtained on bit lines, which are combined together through vertical interface 3521, which effectively adds the signals together in analog form. Optionally, if desired, the combined analog outputs can be digitized elsewhere, such as by converting them to digital form using ADC 3516 (as shown in previous figures).
[0255] Figures 45 to 48 Depicted are examples of structural layout options for die, vertical interfaces, and horizontal interfaces.
[0256] Fig.45A 3D VMM system 4500 is depicted, which includes a first group of dies (die 4501, 4502, 4503, and 4504) arranged in a vertical configuration and connected by vertical interface 3521, and a second group of dies (die 4505, 4506, 4507, and 4508) arranged in a vertical configuration and connected by vertical interface 3521. Here, one die can be connected to two other dies by respective ones of vertical interfaces 3521.
[0257] Fig.46 A 3D VMM system 4600 is depicted, which includes four levels of dies arranged in a vertical staggered configuration, where the first level includes dies 4601 and 4602, the second level includes dies 4603, 4604, and 4605, the third level includes dies 4606 and 4607, and the fourth level includes dies 4608, 4609, and 4610. The dies in different levels are connected to the dies in the upper level and the dies in the lower level through corresponding vertical interfaces 3521. Here, one die can be connected to four other dies through corresponding vertical interfaces 3521.
[0258] Fig.47 A 3D VMM system 4700 is depicted that includes four levels of dies arranged in a vertical staggered configuration, where the first level includes dies 4701 and 4702, the second level includes dies 4703, 4704, and 4705, the third level includes dies 4706 and 4707, and the fourth level includes dies 4708, 4709, and 4710. Dies in different levels are connected by respective vertical interfaces 3521, and dies in the same level are connected by respective horizontal interfaces 3716. Here, one die may be connected to six other dies by respective ones of the vertical interfaces 3521 and by respective ones of the horizontal interfaces 3716.
[0259] Fig.48 An example of a physical layout of a 3D VMM system 4800 is depicted, with only connector 4801 shown. Connector 4801 is located in the die and connects to one or more vertical interfaces 3521 and horizontal interfaces 3716. As can be seen, the die and interfaces can be arranged so that connector 4801 is positioned in a vertically staggered configuration.
[0260] Circuits used in 3D VMM systems
[0261] Figures 49 to 55 Circuitry for use in a 3D VMM system such as those previously described is depicted.
[0262] Fig.49A neuron circuit 4900 is depicted, which may be used in the previously described neuron circuit 3510. Specifically, the neuron circuit 3510 may include an instance of a neuron circuit 4900 for each bit line in the VMM array 3507. The neuron circuit 4900 includes a p-channel metal oxide semiconductor (PMOS) transistor 4901 and an operational amplifier 4902 arranged as shown. One terminal of the PMOS transistor 4901 is attached to a voltage source. The other terminal of the PMOS transistor 4901 is attached to the gate of the PMOS transistor 4910 and the bit line in the VMM array 3507 and the non-inverting terminal of the operational amplifier 4902. The output of the operational amplifier 4902 is attached to the inverting input of the operational amplifier 4902. The current drawn by the bit line I-BL produces a voltage V_IBL output from the operational amplifier 4902. The operational amplifier 4902 acts as a buffer, and V_IBL will maintain its level regardless of the load to which it may be attached. For example, if V_IBL is supplied to the vertical interface 3521, the vertical interface 3521 may have parasitic capacitance or parasitic current. The neuron circuit 4900 will maintain the output voltage V_IBL regardless of changes in the load. Therefore, this is a neuron current buffer circuit. In another embodiment, the neuron current (bit line current) can be amplified or reduced by a current mirror before entering the circuit.
[0263] Fig.50 A neuron circuit 5000 is depicted, which may be used in the previously described neuron circuit 3510. Specifically, the neuron circuit 3510 may include an instance of the neuron circuit 5000 for each bit line in the VMM array 3507. The neuron circuit 5000 includes a controlled switch 5001, a reference memory cell 5002, and an operational amplifier 5003 arranged as shown. The current drawn by the bit line I-BL generates a certain voltage at the non-inverting terminal of the operational amplifier 5003. When the switch 5001 is closed, the feedback VNEUOUT from the output of the operational amplifier 5003 is provided to the control gate terminal of the reference memory cell 5002. Due to the inherent characteristics of the operational amplifier, the operational amplifier 5003 will modify the output voltage VNEUOUT until the voltage on its non-inverting terminal is equal to the voltage VREF on its inverting terminal. It does this by feedback to the control gate terminal of the reference memory cell 5002. Neuron circuit 5000 will maintain output voltage VNEUOUT regardless of the load of parasitic capacitance (such as vertical interface 3521) that may be receiving that voltage. Neuron circuit 5000 uses memory cell 5002 to convert neuron current I-BL to voltage VNEUOUT. Thus, it is a memory cell based current-to-voltage converter using an operational amplifier with feedback.
[0264] Fig.51A force and sense (F / S) driver circuit 5100 is depicted (where the force is caused by an operational amplifier 5103, with input VIN at the positive terminal, and sensing occurs at a target node (node 5101 or 5102), the voltage at the target node is fed back to the negative terminal of the amplifier, and the voltage is made equal to the input voltage VIN by the action of the amplifier), which can be used in the neuron circuit 3510 or elsewhere to accurately deliver the voltage output of the neuron over a varying load. The F / S driver circuit 5100 includes an operational amplifier 5103 and controlled switches 5104, 5105, 5106, and 5107 arranged as shown. Vin is received at the non-inverting input of the operational amplifier 5103, and the output of the operational amplifier 5103 is referred to as VOUT. Switches 5104 and 5106, when closed, connect the output of the operational amplifier 5103 and the inverting input of the operational amplifier 5103, respectively, as VOUTFB to the node 5101 in the first die through the vertical interface 3521. Switches 5105 and 5107, when closed, connect the output of op amp 5103 and the inverting input of op amp 5103, respectively, as VOUTFB to node 5102 in the second die through vertical interface 3521. Driver circuit 5100 provides accurate voltage VOUT to node 5101 and node 5102 regardless of any loading effect caused by vertical interface 3521 or by the first die and the second die. This is because the sense node (drive node) 5101 or 5102 is fed back to the inverting input of op amp 5103.
[0265] Fig.52 Depicted is a differential neuron circuit 5200 that may be used in the previously described neuron circuit 3510. Specifically, the neuron circuit 3510 may include an instance of a differential neuron circuit 5200 for each pair of differential bit lines in the VMM array 3507, such as in an instance where one column stores W+ values and another column stores Wvalues and each pair of W+ and W- represents a stored weight.
[0266] The differential neuron circuit 5200 includes an operational amplifier 5201 configured as shown; variable integrating resistors 5202 and 5203; controlled switches 5204, 5205, 5206, and 5207; and sample and hold (and / or integration) capacitors 5208 and 5209. The differential neuron circuit 5200 receives a differential current BLw+ from the W+ bit line and BLw- from the W- bit line, and outputs voltages Vout+ and Vout-, respectively. The output voltages Vout+ = (BLw+)*R and Vout- = (BLw-)*R, where the variable integrating resistors 5202 and 5203 each have a value equal to R. Once the resistors 5202 and 5203 are removed from the circuit by opening the controlled switches 5206, 5207, and the input current is cut off by opening the controlled switches 5204, 5205, the capacitors 5208 and 5209 act as respective sample and hold (S / H) capacitors to hold the output voltages. A control circuit (not shown) controls the opening and closing of switches 5204, 5205, 5206, and 5207 to provide an integration time. Optionally, the differential output voltages Vout+ and Vout- can be input to an ADC 3516, which converts the differential output voltages Vout+ and Vout- into a set of digital output bits Doutx. Optionally, the circuit can use capacitors 5208 and 5209 as an integrating capacitor to integrate the neuron current to convert the current into a voltage, Vout = time * Ineuron / capacitance. In the case of using an integrating capacitor approach, neuron scaling is provided by variable resistors 5292 and 5203 or a variable integration time and / or variable capacitance.
[0267] Fig.53 A sample-and-hold buffer 5300 is depicted, which may be used in the previously described input circuit 3514. Specifically, the input circuit 3514 may include an instance of a sample-and-hold buffer 5300 for each row in the VMM array 3507. The sample-and-hold buffer 5300 includes a controlled switch 5301, a capacitor 5302, and a buffer 5303. The buffer 5303 may be a unit buffer formed by an operational amplifier. During operation, the switch 5301 is closed, which allows an analog value (e.g., from a DAC 3515) to be stored in the capacitor 5302. The value may then be output from the buffer 5303 driving the row input of the VMM array 3507. For example, the capacitor 5302 may be an actual capacitor, or it may be an inherent capacitor present in the wire.
[0268] Fig.54 A differential sequential address register (SAR) analog-to-digital converter (ADC) 5400 is depicted, which may be used in the ADC 3516 previously described.
[0269] The differential sequential address register analog-to-digital converter 5400 converts an analog input or a differential analog input to a digital output using a binary search through all possible quantization levels to identify the appropriate digital output.
[0270] The differential sequential address register analog-to-digital converter 5400 includes a binary capacitance digital-to-analog converter (CDAC) 5401 , a binary CDAC 5402 (complementary to CDAC 5401 ), a comparator 5403 , and a SAR logic and register 5404 .
[0271] The differential continuous address register analog-to-digital converter 5400 receives differential current inputs Vinp and Vinn. The SAR logic and register 5404 cycles through all possible digital bit combinations, which in turn controls the switches in the CDAC 5401 and 5402 to couple the voltage source to the capacitor. When the output of the comparator 5403 flips, then the digital bit combination in the SAR logic and register 5404 is output as a digital output. Optionally, the SAR logic and register 5404 generates an additional 1-bit digital output DMAJ in the digital output, which is "1" if the majority of the bits in the digital value are "1", and is "0" if the majority of the bits in the corresponding digital value are not "1".
[0272] Fig.55 A reference current based SAR ADC circuit 5500 is depicted, which may be used in the previously described ADC 3516. Specifically, ADC 3516 may include one or more instances of ADC circuit 5500, wherein ADC circuit 5500 converts current I-BL into a digital value Digital Output. ADC circuit 5500 includes a binary current block 5501, a switch 5502, a comparator 5503, and SAR logic and registers 5504. Binary current block 5501 provides a binary reference current to be compared with the bit line input current in a binary search manner (meaning MSB (most significant bit) to LSB (least significant bit) search).
[0273] It should be noted that, as used herein, the terms "above" and "on" both inclusively include "directly on" (no intervening materials, elements, or spaces are provided in between) and "indirectly on" (intervening materials, elements, or spaces are provided in between). Similarly, the term "adjacent" includes "directly adjacent" (no intervening materials, elements, or spaces are provided in between) and "indirectly adjacent" (intervening materials, elements, or spaces are provided in between), "mounted to" includes "directly mounted to" (no intervening materials, elements, or spaces are provided in between) and "indirectly mounted to" (intervening materials, elements, or spaces are provided in between), and "electrically coupled to" includes "directly electrically coupled to" (no intervening materials or elements electrically connecting the elements together) and "indirectly electrically coupled to" (intervening materials or elements electrically connecting the elements together). For example, forming an element "above" a substrate may include directly forming the element on the substrate without intervening materials / elements in between, and indirectly forming the element on the substrate with one or more intervening materials / elements in between.
Claims
1. A three-dimensional integrated circuit for use in an artificial neural network, comprising: a first die, the first die comprising a first vector-matrix multiplication array and a first input multiplexer, the first die being located on a first vertical layer; a second die, the second die comprising input circuitry, the second die being located on a second vertical layer different from the first vertical layer; and one or more vertical interfaces coupling the first die and the second die; Wherein during a read operation, the input circuit provides an input signal to the first input multiplexer through at least one of the one or more vertical interfaces, the first input multiplexer applies the input signal to one or more rows in the first vector-matrix multiplication array, and the first vector-matrix multiplication array generates an output. 2 . The three-dimensional integrated circuit of claim 1 , wherein the second die comprises a digital-to-analog converter for converting a digital input into an analog input provided as the input signal to the input circuit. 3 . The three-dimensional integrated circuit of claim 1 , wherein the first die comprises a neuron circuit configured to buffer the output. 4 . The three-dimensional integrated circuit of claim 1 , wherein the first die comprises a column multiplexer configured to send the output to the second die or the third die through at least one of the one or more vertical interfaces.
5. The three-dimensional integrated circuit according to claim 1, comprising: A third die includes an analog-to-digital converter for converting the output from the first die into a digital output, the third die being located on a third vertical layer different from the first vertical layer and the second vertical layer.
6. The three-dimensional integrated circuit according to claim 5, comprising: A fourth die includes a high voltage generator, an analog circuit, and a temperature compensation circuit, and the fourth die is located on a fourth vertical layer different from the first vertical layer, the second vertical layer, and the third vertical layer.
7. The three-dimensional integrated circuit according to claim 6, comprising: A fifth die, comprising a second vector-matrix multiplication array, a second input multiplexer, a high voltage decoder, and a neuron circuit, the fifth die being located on a fifth vertical layer different from the first vertical layer, the second vertical layer, the third vertical layer, and the fourth vertical layer.
8. The three-dimensional integrated circuit according to claim 1, comprising: A third die includes a second vector-matrix multiplication array and a second input multiplexer, the third die being located on the first vertical layer.
9. The three-dimensional integrated circuit of claim 1, wherein the vector-matrix multiplication array comprises a plurality of non-volatile memory cells.
10. The three-dimensional integrated circuit of claim 9, wherein the plurality of non-volatile memory cells comprise stacked gate flash memory cells.
11. The three-dimensional integrated circuit of claim 9, wherein the plurality of non-volatile memory cells comprise split-gate flash memory cells.
12. A method comprising: providing an input signal from an input circuit located on the first die to an input multiplexer located on the second die through one or more vertical interfaces; applying the input signal to one or more rows in a neural network array by the input multiplexer; as well as generating an output by the neural network array; The first die and the second die are located on different vertical layers.
13. The method of claim 12, wherein the neural network array comprises a plurality of non-volatile memory cells.
14. The method of claim 13, wherein the plurality of non-volatile memory cells comprises stacked gate flash memory cells.
15. The method of claim 13, wherein the plurality of non-volatile memory cells comprise split-gate flash memory cells.
16. An apparatus comprising: a first die, the first die comprising a first vector-matrix multiplication array, the first vector-matrix multiplication array comprising a plurality of non-volatile memory cells arranged in rows and columns, the first die being located on a first vertical layer; a second die, the second die comprising a second vector-matrix multiplication array, the second vector-matrix multiplication array comprising a plurality of non-volatile memory cells arranged in rows and columns, the second die being located on a second vertical layer different from the first vertical layer; and one or more vertical interfaces coupling the first die and the second die; Wherein during a programming operation, one or more non-volatile memory cells in the first array are capable of storing i bits and one or more non-volatile memory cells in the second array are capable of storing, bits, where i≠j. 17 . The apparatus of claim 16 , wherein the first die is fabricated according to a first semiconductor process, and the second die is fabricated according to a second semiconductor process different from the first semiconductor process.
18. The apparatus according to claim 16, comprising: a first set of bit lines coupled to the first vector-matrix multiplication array; and A second group of bit lines different from the first group of bit lines is coupled to the second vector-matrix multiplication array.
19. The apparatus according to claim 18, comprising: a first set of control gate lines coupled to the first vector-matrix multiplication array; and A second group of control gate lines different from the first group of control gate lines is coupled to the second vector-matrix multiplication array.
20. The device of claim 19, wherein the first set of control gate lines are coupled to the second set of control gate lines through respective vertical interfaces.
21. The apparatus of claim 18, wherein the first set of bit lines are coupled to the second set of bit lines through respective vertical interfaces.
22. The apparatus according to claim 21, comprising: a first set of control gate lines coupled to the first vector-matrix multiplication array; and A second set of control gate lines is coupled to the second vector-matrix multiplication array.
23. The device of claim 22, wherein the first set of control gate lines are coupled to the second set of control gate lines through respective vertical interfaces.
24. The apparatus of claim 16, wherein the plurality of nonvolatile memory cells in the first die and the plurality of nonvolatile memory cells in the second die each comprise stacked gate flash memory cells.
25. The apparatus of claim 16, wherein the plurality of nonvolatile memory cells in the first die and the plurality of nonvolatile memory cells in the second die each comprise a split-gate flash memory cell.
26. A method comprising: In a first neural network array in a first die, storing a value comprising i bits in a non-volatile memory cell; as well as In a second neural network array in a second die, storing a value comprising i bits in a non-volatile memory cell, where i≠y; The first die and the second die are located on different vertical layers.
27. An apparatus comprising: a first die, the first die comprising a first vector-matrix multiplication array, the first vector-matrix multiplication array comprising a respective plurality of non-volatile memory cells arranged in rows and columns, the first die being located on a first vertical layer; a second die comprising a second vector-matrix multiplication array including a respective plurality of nonvolatile memory cells arranged in rows and columns, the second die being located on a second vertical layer different from the first layer; and A third die includes one or more of a digital-to-analog converter and an analog-to-digital converter.
28. The apparatus of claim 27, wherein the analog-to-digital converter is a capacitor-based successive approximation register analog-to-digital converter.
29. The apparatus of claim 27, wherein the analog-to-digital converter is a successive approximation register analog-to-digital converter based on a reference current.
30. The apparatus of claim 27, wherein the respective plurality of non-volatile memory cells in the first vector-matrix multiplication array and the second vector-matrix multiplication array comprise stacked gate flash memory cells.
31. The apparatus of claim 27, wherein the respective plurality of non-volatile memory cells in the first vector-matrix multiplication array and the second vector-matrix multiplication array comprise split-gate flash memory cells.
32. An apparatus comprising: a first die, the first die comprising a first vector-matrix multiplication array, the first vector-matrix multiplication array comprising a plurality of non-volatile memory cells arranged in rows and columns, the first die being located on a first vertical layer; a second die, the second die comprising a second vector-matrix multiplication array, the second vector-matrix multiplication array comprising a plurality of non-volatile memory cells arranged in rows and columns, the second die being located on a second vertical layer different from the first vertical layer; and A third die includes a neuron circuit.
33. The apparatus of claim 32, wherein the neuron circuit comprises: a p-channel metal oxide semiconductor transistor comprising a first terminal coupled to a voltage source, a gate, and a second terminal coupled to the gate and the neuron; and An operational amplifier includes a non-inverting input coupled to the second terminal and the gate of the p-channel metal oxide semiconductor transistor, an inverting input, and an output coupled to the inverting input to generate a voltage output in response to current from the neuron.
34. The apparatus of claim 32, wherein the neuron circuit comprises: switch; a reference memory cell including a bit line terminal, a source line terminal, and a control gate terminal coupled to the neuron; as well as An operational amplifier includes an inverting input coupled to a reference voltage, a non-inverting input coupled to the bit line terminal of the reference memory cell, and an output switchably coupled to the control gate terminal of the reference memory cell through the switch.
35. The apparatus of claim 32, wherein the neuron circuit comprises: an operational amplifier including an inverting input, a non-inverting input, and a first output coupled to the first output node and a second output coupled to the second output node; a first variable integrating resistor switchably coupled between the first output node and the inverting input of the operational amplifier via a first switch; a second variable integrating resistor switchably coupled between the second output node and the non-inverting input of the operational amplifier via a second switch; a first capacitor switchably coupled between a first input current from a bit line through a third switch and the first output node; and A second capacitor is switchably coupled between a second input current from the bit line through a fourth switch and the second output node.
36. The apparatus of claim 35, wherein the first input current and the second input current are differential current signals, and the first output node and the second output node comprise differential voltage signals.
37. The apparatus of claim 36, wherein the first input current is received from a W+ bit line and the second input current is received from a W- bit line.
38. The apparatus of claim 36, comprising: An analog-to-digital converter is configured to convert the differential voltage signal into a set of digital output bits.
39. The apparatus of claim 32, wherein the plurality of non-volatile memory cells in the first vector-matrix multiplication array and the second vector-matrix multiplication array comprise stacked gate flash memory cells.
40. The apparatus of claim 32, wherein the plurality of non-volatile memory cells in the first vector-matrix multiplication array and the second vector-matrix multiplication array comprise split-gate flash memory cells.
41. An apparatus comprising: a first die, the first die comprising a first vector-matrix multiplication array, the first vector-matrix multiplication array comprising a respective plurality of non-volatile memory cells arranged in rows and columns, the first die being located on a first vertical layer; a second die comprising a second vector-matrix multiplication array including a respective plurality of nonvolatile memory cells arranged in rows and columns, the second die being located on a second vertical layer different from the first vertical layer; and A third die includes digital circuits including one or more of a microcontroller, digital logic, or a single instruction multiple data processor.
42. The apparatus of claim 41, wherein the third die comprises a digital accelerator.
43. The apparatus of claim 41, wherein the third die comprises static random access memory.
44. The apparatus of claim 41, wherein the third die comprises physical input / output connections.
45. The apparatus of claim 41, wherein the third die comprises a register.
46. The apparatus of claim 41, wherein the corresponding plurality of non-volatile memory cells in the first vector-matrix multiplication array and the corresponding plurality of non-volatile memory cells in the second vector-matrix multiplication array comprise stacked gate flash memory cells.
47. The apparatus of claim 41, wherein the corresponding plurality of non-volatile memory cells in the first vector-matrix multiplication array and the corresponding plurality of non-volatile memory cells in the second vector-matrix multiplication array comprise split-gate flash memory cells.
48. An apparatus comprising: a first vertical layer, the first vertical layer comprising a first vector-matrix multiplication array comprising a corresponding plurality of non-volatile memory cells arranged in rows and columns and a second vector-matrix multiplication array comprising a corresponding plurality of non-volatile memory cells arranged in rows and columns; one or more corresponding horizontal interfaces, the one or more corresponding horizontal interfaces coupling the first vector-matrix multiplication array and the second vector-matrix multiplication array; a second vertical layer, the second vertical layer comprising a third vector-matrix multiplication array comprising a corresponding plurality of non-volatile memory cells arranged in rows and columns and a fourth vector-matrix multiplication array comprising a corresponding plurality of non-volatile memory cells arranged in rows and columns; and One or more corresponding horizontal interfaces, the one or more corresponding horizontal interfaces coupling the third vector-matrix multiplication array and the fourth vector-matrix multiplication array.
49. The apparatus of claim 48, wherein the first vector-matrix multiplication array is located on a first die, the second vector-matrix multiplication array is located on a second die, the third vector-matrix multiplication array is located on a third die, and the fourth vector-matrix multiplication array is located on a fourth die.
50. The apparatus of claim 49, wherein the first die and the third die are vertically aligned, and the second die and the fourth die are vertically aligned.
51. The device of claim 49, wherein the first die and the second die are vertically staggered with the third die and the fourth die.
52. The apparatus of claim 48, wherein the respective plurality of non-volatile memory cells in the first vector-matrix multiplication array, the second vector-matrix multiplication array, the third vector-matrix multiplication array, and the fourth vector-matrix multiplication array comprise stacked gate flash memory cells.
53. The apparatus of claim 48, wherein the corresponding plurality of non-volatile memory cells in the first vector-matrix multiplication array, the second vector-matrix multiplication array, the third vector-matrix multiplication array, and the fourth vector-matrix multiplication array comprise split-gate flash memory cells.
Citation Information
Patent Citations
Method and apparatus for programming analog neural memory in a deep learning artificial neural network
CN111837190A
Deep Learning Neural Network Classifier Using Non-volatile Memory Array
US20170337466A1
Vector-by-matrix multiplier modules based on non-volatile 2d and 3D memory arrays
US20190213234A1
System for converting neuron current into neuron current-based time pulses in an analog neural memory in a deep learning artificial neural network
US20200234111A1
