Precise programming method and apparatus for analog neural memory in an artificial neural network
11755899 · 2023-09-12
Assignee
Inventors
- Hieu Van Tran (San Jose, CA)
- Steven Lemke (Boulder Creek, CA)
- Vipin Tiwari (Dublin, CA)
- Nhan Do (Saratoga, CA)
- Mark Reiten (Alamo, CA)
Cpc classification
G11C16/0425
PHYSICS
G11C2216/04
PHYSICS
G11C16/3459
PHYSICS
G06F17/16
PHYSICS
G11C16/14
PHYSICS
International classification
G06F17/16
PHYSICS
G11C11/56
PHYSICS
G11C16/14
PHYSICS
Abstract
Numerous embodiments of a precision programming algorithm and apparatus are disclosed for precisely and quickly depositing the correct amount of charge on the floating gate of a non-volatile memory cell within a vector-by-matrix multiplication (VMM) array in an artificial neural network. Selected cells thereby can be programmed with extreme precision to hold one of N different values.
Claims
1. A method comprising: performing a first programming process comprising multiple program-verify cycles, wherein a programming voltage of increasing magnitude is applied to a terminal of a selected non-volatile memory cell in program-verify cycles after the first program-verify cycle, wherein each program-verify cycle comprises: applying a first voltage to one of an erase gate terminal and a control gate terminal of the selected non-volatile memory cell; measuring a first current that results through the selected non-volatile memory cell; applying a second voltage to the one of the erase gate terminal and the control gate terminal of the selected non-volatile memory cell; measuring a second current that results through the selected non-volatile memory cell; determining a slope value based on the first voltage, the second voltage, the first current, and the second current; and determining a next programming voltage of the programming voltages of increasing magnitude for the next program-verify cycle based on the determined slope value.
2. The method of claim 1, further comprising: programming the selected non-volatile memory cell using the next programming voltage.
3. The method of claim 2, further comprising: repeating the steps of determining a next programming voltage and programming the non-volatile memory cell using the next programming voltage until a current through the selected non-volatile memory cell during a read or verify operation is less than or equal to a first threshold current value.
4. A method comprising: performing a first programming process comprising multiple program-verify cycles, wherein a programming voltage of increasing magnitude is applied to a terminal of a selected non-volatile memory cell in each program-verify cycle after the first program-verify cycle, wherein the step of performing a first programming process further comprises: when a current through the selected non-volatile memory cell is less than or equal to a third threshold current value, erasing the selected non-volatile memory cell and repeating the first programming process.
5. The method of claim 4, further comprising: performing a third programming process until a current through the selected non-volatile memory cell during a read or verify operation is less than or equal to a fourth threshold current value.
6. The method of claim 4, wherein the selected non-volatile memory cell is a split-gate flash memory cell.
7. The method of claim 4, wherein the selected non-volatile memory cell is in a vector-by-matrix multiplication array in an analog neural network.
8. A method comprising: performing a first programming process comprising multiple program-verify cycles, wherein a programming voltage of increasing magnitude is applied to a terminal of a selected non-volatile memory cell in each program-verify cycle after the first program-verify cycle and wherein each program-verify cycle comprises verifying that a current through the selected non-volatile memory cell during a read or verify operation is less than or equal to a first threshold current value; wherein when the current through the selected non-volatile memory cell during the read or verify operation is less than or equal to the first threshold current value, performing a second programming process until a current through the selected non-volatile memory cell during a read or verify operation is less than or equal to a second threshold current value, wherein the second programming process comprises applying voltage pulses of increasing magnitude to a control gate of the selected non-volatile memory cell, wherein the second programming process further comprises applying voltage pulses of decreasing magnitude to an erase gate of the selected non-volatile memory cell.
9. A method of programming a selected non-volatile memory cell to store one of N possible values, where N is an integer greater than 2, the selected non-volatile memory cell comprising a floating gate, a control gate terminal, an erase gate terminal, and a source line terminal, the method comprising: performing a first programming process comprising multiple program-verify cycles, wherein a first programming voltage of increasing magnitude is applied to the control gate of the selected non-volatile memory cell and a second programming voltage of decreasing magnitude is applied to the erase gate of the selected non-volatile memory cell during the programming process.
10. The method of claim 9, wherein each program-verify cycle comprises verifying that a current through the selected non-volatile memory cell during a read or verify operation is less than or equal to a first threshold current value.
11. The method of claim 9, wherein each program-verify cycle comprises: applying a first voltage to one of the erase gate and the control gate of the selected non-volatile memory cell; measuring a first current that results through the selected non-volatile memory cell; applying a second voltage to the one of the erase gate and the control gate of the selected non-volatile memory cell; measuring a second current that results through the selected non-volatile memory cell; determining a slope value based on the first voltage, the second voltage, the first current, and the second current; determining a next programming voltage of the programming voltages of increasing and decreasing magnitude, respectively, based on the slope value; programming the non-volatile memory cell with the next programming voltage; repeating the steps of determining a next programming voltage and programming the non-volatile memory cell with the next programming voltage until a current through the selected non-volatile memory cell during a read or verify operation is less than or equal to a first threshold current value.
12. The method of claim 9, further comprising: performing a second programming process until a current through the selected non-volatile memory cell during a read or verify operation is less than or equal to a second threshold current value.
13. The method of claim 12, further comprising: performing a third programming process until a current through the selected non-volatile memory cell during a read or verify operation is less than or equal to a fourth threshold current value.
14. The method of claim 9, wherein the step of performing a first programming process further comprises: when a current through the selected non-volatile memory cell is less than or equal to a third threshold current value, erasing the selected non-volatile memory cell and repeating the first programming process.
15. A method of programming a selected non-volatile memory cell to store one of N possible values, where N is an integer greater than 2, the selected non-volatile memory cell comprising a first gate, a first terminal, and a second terminal, the method comprising: performing a plurality of program-verify cycles on a selected split-gate memory cell comprising a source line terminal, a control gate terminal, and a floating gate, wherein during each of the program-verify cycles a first programming voltage is applied to the source line terminal and the first programming voltage decreases with each subsequent cycle and a second programming voltage is applied to the control gate terminal and the second programming voltage increases with each subsequent cycle.
Description
BRIEF DESCRIPTION OF THE DRAWINGS
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
(10)
(11)
(12)
(13)
(14)
(15)
(16)
(17)
(18)
(19)
(20)
(21)
(22)
(23)
(24)
(25)
(26)
(27)
(28)
(29)
(30)
(31)
(32)
(33)
(34)
(35)
(36)
(37)
(38)
(39)
(40)
(41)
(42)
(43)
(44)
(45)
(46)
(47)
(48)
(49)
(50)
(51)
(52)
(53)
(54)
(55)
(56)
(57)
(58)
(59)
(60)
(61)
(62)
DETAILED DESCRIPTION OF THE INVENTION
(63) The artificial neural networks of the present invention utilize a combination of CMOS technology and non-volatile memory arrays.
(64) Non-Volatile Memory Cells
(65) Digital non-volatile memories are 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 non-volatile memory cells, which are a type of flash memory cells. Such a memory cell 210 is shown in
(66) Memory cell 210 is erased (where electrons are removed from the floating gate) by placing a high positive voltage on the word line terminal 22, which causes electrons on the floating gate 20 to tunnel through the intermediate insulation from the floating gate 20 to the word line terminal 22 via Fowler-Nordheim tunneling.
(67) Memory cell 210 is programmed (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. Electron current will flow from the source region 14 (source line terminal) towards the drain region 16. The electrons will accelerate and become heated when they reach the gap between the word line terminal 22 and the floating gate 20. Some of the heated electrons will be injected through the gate oxide onto the floating gate 20 due to the attractive electrostatic force from the floating gate 20.
(68) Memory cell 210 is read by placing positive read voltages on the drain region 16 and word line terminal 22 (which turns on the portion of the channel region 18 under the word line terminal). If the floating gate 20 is positively charged (i.e. erased of electrons), then the portion of the channel region 18 under the floating gate 20 is turned on as well, and current will flow across the channel region 18, which is sensed as the erased or “1” state. If the floating gate 20 is negatively charged (i.e. programmed with electrons), then the portion of the channel region under the floating gate 20 is mostly or entirely turned off, and current will not flow (or there will be little flow) across the channel region 18, which is sensed as the programmed or “0” state.
(69) Table No. 1 depicts typical voltage ranges that can be applied to the terminals of memory cell 110 for performing read, erase, and program operations:
(70) TABLE-US-00001 TABLE NO. 1 Operation of Flash Memory Cell 210 of FIG. 2 WL BL SL Read 1 0.5-3 V 0.1-2 V 0 V Read 2 0.5-3 V 0-2 V 2-0.1 V Erase ~11-13 V 0 V 0 V Program 1-2 V 1-3 μA 9-10 V “Read 1” is a read mode in which the cell current is output on the bit line. “Read 2” is a read mode in which the cell current is output on the source line terminal.
(71)
(72)
(73) Table No. 2 depicts typical voltage ranges that can be applied to the terminals of memory cell 410 for performing read, erase, and program operations:
(74) TABLE-US-00002 TABLE NO. 2 Operation of Flash Memory Cell 410 of FIG. 4 WL/SG BL CG EG SL Read 1 0.5-2 V 0.1-2 V 0-2.6 V 0-2.6 V 0 V Read 2 0.5-2 V 0-2 V 0-2.6 V 0-2.6 V 2-0.1 V Erase −0.5 V/0 V 0 V 0 V/−8 V 8-12 V 0 V Program 1 V 1 μA 8-11 V 4.5-9 V 4.5-5 V “Read 1” is a read mode in which the cell current is output on the bit line. “Read 2” is a read mode in which the cell current is output on the source line terminal.
(75)
(76)
(77) Table No. 3 depicts typical voltage ranges that can be applied to the terminals of memory cell 610 for performing read, erase, and program operations:
(78) TABLE-US-00003 TABLE NO. 3 Operation of Flash Memory Cell 610 of FIG. 6 WL/SG BL EG SL Read 1 0.5-2.2 V 0.1-2 V 0-2.6 V 0 V Read 2 0.5-2.2 V 0-2 V 0-2.6 V 2-0.1 V Erase −0.5 V/0 V 0 V 11.5 V 0 V Program 1 V 2-3 μA 4.5 V 7-9 V “Read 1” is a read mode in which the cell current is output on the bit line. “Read 2” is a read mode in which the cell current is output on the source line terminal.
(79)
(80) Table No. 4 depicts typical voltage ranges that can be applied to the terminals of memory cell 710 and substrate 12 for performing read, erase, and program operations:
(81) TABLE-US-00004 TABLE NO. 4 Operation of Flash Memory Cell 710 of FIG. 7 CG BL SL Substrate Read 1 0-5 V 0.1-2 V 0-2 V 0 V Read 2 0.5-2 V 0-2 V 2-0.1 V 0 V Erase −8 to −10 V/0 V FLT FLT 8-10 V/15-20 V Program 8-12 V 3-5 V/0 V 0 V/3-5 V 0 V “Read 1” is a read mode in which the cell current is output on the bit line. “Read 2” is a read mode in which the cell current is output on the source line terminal.
Optionally, in arrays comprising rows and columns of memory cells 210, 310, 410, 510, 610, or 710, source lines can be coupled to one row of memory cells or to two adjacent rows of memory cells. That is, source line terminals can be shared by adjacent rows of memory cells.
(82) In order to utilize the memory arrays comprising one of the types of non-volatile memory cells described above in an artificial neural network, two modifications are made. First, the lines are configured so that each memory cell can be individually programmed, erased, and read without adversely affecting the memory state of other memory cells in the array, as further explained below. Second, continuous (analog) programming of the memory cells is provided.
(83) Specifically, the memory state (i.e. charge on the floating gate) of each memory cell in the array can be continuously changed from a fully erased state to a fully programmed state, independently and with minimal disturbance of other memory cells. In another embodiment, the memory state (i.e., charge on the floating gate) of each memory cell in the array can be continuously changed from a fully programmed state to a fully erased state, and vice-versa, independently and with minimal disturbance of other memory cells. This means the cell storage is analog or at the very least can store one of many discrete values (such as 16 or 64 different values), which allows for very precise and individual tuning of all the cells in the memory array, and which makes the memory array ideal for storing and making fine tuning adjustments to the synapsis weights of the neural network.
(84) The methods and means described herein may apply to other non-volatile memory technologies such as SONOS (silicon-oxide-nitride-oxide-silicon, charge trap in nitride), MONOS (metal-oxide-nitride-oxide-silicon, metal charge trap in nitride), ReRAM (resistive ram), PCM (phase change memory), MRAM (magnetic ram), FeRAM (ferroelectric ram), OTP (bi-level or multi-level one time programmable), and CeRAM (correlated electron ram), without limitation. The methods and means described herein may apply to volatile memory technologies used for neural network such as SRAM, DRAM, and other volatile synapse cells, without limitation.
(85) Neural Networks Employing Non-Volatile Memory Cell Arrays
(86)
(87) S0 is the input layer, which for this example 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, each pixel being 5 bit precision). The synapses CB1 going from input layer S0 to layer C1 apply different sets of weights in some instances and shared weights in other instances, and scan the input image with 3×3 pixel overlapping filters (kernel), shifting the filter by 1 pixel (or more than 1 pixel as dictated by the model). Specifically, values for 9 pixels in a 3×3 portion of the image (i.e., referred to as a filter or kernel) are provided to the synapses CB1, where these 9 input values are multiplied by the appropriate weights and, after summing the outputs of that multiplication, a single output value is determined and provided by a first synapse of CB1 for generating a pixel of one of the layers of feature map C1. The 3×3 filter is then shifted one pixel to the right within input layer S0 (i.e., adding the column of three pixels on the right, and dropping the column of three pixels on the left), whereby the 9 pixel values in this newly positioned filter are provided to the synapses CB1, where they are multiplied by the same weights and a second single output value is determined by the associated synapse. This process is continued until the 3×3 filter scans across the entire 32×32 pixel image of input layer S0, for all three colors and for all bits (precision values). The process is then repeated using different sets of weights to generate a different feature map of C1, until all the features maps of layer C1 have been calculated.
(88) In layer C1, in the present example, there are 16 feature maps, with 30×30 pixels each. Each pixel is a new feature pixel extracted from multiplying the inputs and kernel, and therefore each feature map is a two dimensional array, and thus in this example layer C1 constitutes 16 layers of two dimensional arrays (keeping in mind that the layers and arrays referenced herein are logical relationships, not necessarily physical relationships—i.e., the arrays are not necessarily oriented in physical two dimensional arrays). Each of the 16 feature maps in layer C1 is generated by one of sixteen different sets of synapse weights applied to the filter scans. The C1 feature maps could all be directed to different aspects of the same image feature, such as boundary identification. For example, the first map (generated using a first weight set, shared for all scans used to generate this first map) could identify circular edges, the second map (generated using a second weight set different from the first weight set) could identify rectangular edges, or the aspect ratio of certain features, and so on.
(89) An activation function P1 (pooling) is applied before going from layer C1 to layer S1, which pools values from consecutive, non-overlapping 2×2 regions in each feature map. The purpose of the pooling function is to average out the nearby location (or a max function can also be used), to reduce the dependence of the edge location for example and to reduce the data size before going to the next stage. At layer S1, there are 16 15×15 feature maps (i.e., sixteen different arrays of 15×15 pixels each). The synapses CB2 going from layer S1 to layer C2 scan maps in S1 with 4×4 filters, with a filter shift of 1 pixel. At layer C2, there are 22 12×12 feature maps. An activation function P2 (pooling) is applied before going from layer C2 to layer S2, which pools 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 at the synapses CB3 going from layer S2 to layer C3, where every neuron in layer C3 connects to every map in layer S2 via a respective synapse of CB3. At layer C3, there are 64 neurons. The synapses CB4 going from layer C3 to the output layer S3 fully connects C3 to S3, i.e. every neuron in layer C3 is connected to every neuron in layer S3. The output at S3 includes 10 neurons, where the highest output neuron determines the class. This output could, for example, be indicative of an identification or classification of the contents of the original image.
(90) Each layer of synapses is implemented using an array, or a portion of an array, of non-volatile memory cells.
(91)
(92) VMM array 33 serves two purposes. First, it stores the weights that will be used by the VMM system 32. Second, VMM array 33 effectively multiplies the inputs by the weights stored in VMM array 33 and adds them up per output line (source line or bit line) to produce the output, which will be the input to the next layer or input to the final layer. By performing the multiplication and addition function, VMM array 33 negates the need for separate multiplication and addition logic circuits and is also power efficient due to its in-situ memory computation.
(93) The output of VMM array 33 is supplied to a differential summer (such as a summing op-amp or a summing current mirror) 38, which sums up the outputs of VMM array 33 to create a single value for that convolution. The differential summer 38 is arranged to perform summation of both positive weight and negative weight inputs to output the single value.
(94) The summed up output values of differential summer 38 are then supplied to an activation function circuit 39, which rectifies the output. The activation function circuit 39 may provide sigmoid, tanh, ReLU functions, or any other non-linear function. The rectified output values of activation function circuit 39 become an element of a feature map of the next layer (e.g. C1 in
(95) The input to VMM system 32 in
(96)
(97) The output generated by input VMM system 32a is provided as an input to the next VMM system (hidden level 1) 32b, which in turn generates an output that is provided as an input to the next VMM system (hidden level 2) 32c, and so on. The various layers of VMM system 32 function as different layers of synapses and neurons of a convolutional neural network (CNN). Each VMM system 32a, 32b, 32c, 32d, and 32e can be a stand-alone, physical system comprising a respective non-volatile memory array, or multiple VMM systems could utilize different portions of the same physical non-volatile memory array, or multiple VMM systems could utilize overlapping portions of the same physical non-volatile memory array. Each VMM system 32a, 32b, 32c, 32d, and 32e can also be time multiplexed for various portion of its array or neurons. The example shown in
(98) VMM Arrays
(99)
(100) In VMM array 1100, control gate lines, such as control gate line 1103, run in a vertical direction (hence reference array 1102 in the row direction is orthogonal to control gate line 1103), and erase gate lines, such as erase gate line 1104, run in a horizontal direction. Here, the inputs to VMM array 1100 are provided on the control gate lines (CG0, CG1, CG2, CG3), and the output of VMM array 1100 emerges on the source lines (SL0, SL1). In one embodiment, only even rows are used, and in another embodiment, only odd rows are used. The current placed on each source line (SL0, SL1, respectively) performs a summing function of all the currents from the memory cells connected to that particular source line.
(101) As described herein for neural networks, the non-volatile memory cells of VMM array 1100, i.e. the flash memory of VMM array 1100, are preferably configured to operate in a sub-threshold region.
(102) The non-volatile reference memory cells and the non-volatile memory cells described herein are biased in weak inversion:
Ids=Io*e.sup.Vg−Vth/nVt=w*Io*e.sup.(Vg)/nVt,
where w=e.sup.(−Vth)/nVt
where Ids is the drain to source current; Vg is gate voltage on the memory cell; Vth is threshold voltage of the memory cell; Vt is thermal voltage=k*T/q with k being the Boltzmann constant, T the temperature in Kelvin, and q the electronic charge; n is a slope factor=1+(Cdep/Cox) with Cdep=capacitance of the depletion layer, and Cox capacitance of the gate oxide layer; Io is the memory cell current at gate voltage equal to threshold voltage, Io is proportional to (Wt/L)*u*Cox*(n−1)*Vt.sup.2 where u is carrier mobility and Wt and L are width and length, respectively, of the memory cell.
(103) For an I-to-V log converter using a memory cell (such as a reference memory cell or a peripheral memory cell) or a transistor to convert input current Ids, into an input voltage, Vg:
Vg=n*Vt*log [Ids/wp*Io]
Here, wp is w of a reference or peripheral memory cell.
(104) For an I-to-V log converter using a memory cell (such as a reference memory cell or a peripheral memory cell) or a transistor to convert input current Ids, into an input voltage, Vg:
Vg=n*Vt*log [Ids/wp*Io]
(105) Here, wp is w of a reference or peripheral memory cell.
(106) For a memory array used as a vector matrix multiplier VMM array, the output current is:
Iout=wa*Io*e.sup.(Vg)/nVt namely
Iout=(wa/wp)*Iin=W*Iin
W=e.sup.(Vthp−Vtha)/nVt
Iin=wp*Io*e.sup.(Vg)/nVt Here, wa=w of each memory cell in the memory array.
(107) A wordline or control gate can be used as the input for the memory cell for the input voltage.
(108) Alternatively, the non-volatile memory cells of VMM arrays described herein can be configured to operate in the linear region:
Ids=beta*(Vgs−Vth)*Vds;beta=u*Cox*Wt/L,
Wα(Vgs−Vth), meaning weight W in the linear region is proportional to (Vgs−Vth)
(109) A wordline or control gate or bitline or sourceline can be used as the input for the memory cell operated in the linear region. The bitline or sourceline can be used as the output for the memory cell.
(110) For an I-to-V linear converter, a memory cell (such as a reference memory cell or a peripheral memory cell) or a transistor operating in the linear region or a resistor can be used to linearly convert an input/output current into an input/output voltage.
(111) Alternatively, the memory cells of VMM arrays described herein can be configured to operate in the saturation region:
Ids=½*beta*(Vgs−Vth).sup.2;beta=u*Cox*Wt/L Wα (Vgs−Vth).sup.2, meaning weight W is proportional to (Vgs−Vth).sup.2
(112) A wordline, control gate, or erase gate can be used as the input for the memory cell operated in the saturation region. The bitline or sourceline can be used as the output for the output neuron.
(113) Alternatively, the memory cells of VMM arrays described herein can be used in all regions or a combination thereof (sub threshold, linear, or saturation).
(114) Other embodiments for VMM array 33 of
(115)
(116) Memory array 1203 serves two purposes. First, it stores the weights that will be used by the VMM array 1200 on respective memory cells thereof. Second, memory array 1203 effectively multiplies the inputs (i.e. current inputs provided in terminals BLR0, BLR1, BLR2, and BLR3, which reference arrays 1201 and 1202 convert into the input voltages to supply to wordlines WL0, WL1, WL2, and WL3) by the weights stored in the memory array 1203 and then adds all the results (memory cell currents) to produce the output on the respective bit lines (BL0-BLN), which will be the input to the next layer or input to the final layer. By performing the multiplication and addition function, memory array 1203 negates the need for separate multiplication and addition logic circuits and is also power efficient. Here, the voltage inputs are provided on the word lines WL0, WL1, WL2, and WL3, and the output emerges on the respective 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 bitline.
(117) Table No. 5 depicts operating voltages for VMM array 1200. The columns in the table indicate the voltages placed on word lines for selected cells, word lines for unselected cells, bit lines for selected cells, bit lines for unselected cells, source lines for selected cells, and source lines for unselected cells, where FLT indicates floating, i.e. no voltage is imposed. The rows indicate the operations of read, erase, and program.
(118) TABLE-US-00005 TABLE NO. 5 Operation of VMM Array 1200 of FIG. 12: WL WL -unsel BL BL -unsel SL SL -unsel Read 0.5-3.5 V −0.5 V/0 V 0.1-2 V 0.6 V-2 V/FLT 0 V 0 V (Ineuron) Erase ~5-13 V 0 V 0 V 0 V 0 V 0 V Program 1-2 V −0.5 V/0 V 0.1-3 uA Vinh ~2.5 V 4-10 V 0-1 V/FLT
(119)
(120) Table No. 6 depicts operating voltages for VMM array 1300. The columns in the table indicate the voltages placed on word lines for selected cells, word lines for unselected cells, bit lines for selected cells, bit lines for unselected cells, source lines for selected cells, and source lines for unselected cells. The rows indicate the operations of read, erase, and program.
(121) TABLE-US-00006 TABLE NO. 6 Operation of VMM Array 1300 of FIG. 13 WL WL -unsel BL BL -unsel SL SL -unsel Read 0.5-3.5 V −0.5 V/0 V 0.1-2 V 0.1 V-2 V/FLT ~0.3-1 V 0 V (Ineuron) Erase ~5-13 V 0 V 0 V 0 V 0 V SL-inhibit (~4-8 V) Program 1-2 V −0.5 V/0 V 0.1-3 uA Vinh ~2.5 V 4-10 V 0-1 V/FLT
(122)
(123) Memory array 1403 serves two purposes. First, it stores the weights that will be used by the VMM array 1400. Second, memory array 1403 effectively multiplies the inputs (current inputs provided to terminals BLR0, BLR1, BLR2, and BLR3, for which reference arrays 1401 and 1402 convert these current inputs into the input voltages to supply to the control gates (CG0, CG1, CG2, and CG3) by the weights stored in the memory array and then add all the results (cell currents) to produce the output, which appears on BL0-BLN, and will be the input to the next layer or input to the final layer. By performing the multiplication and addition function, the memory array negates the need for separate multiplication and addition logic circuits and is also power efficient. Here, the inputs are provided on the control gate lines (CG0, CG1, CG2, and CG3), and the output emerges on the bitlines (BL0-BLN) during a read operation. The current placed on each bitline performs a summing function of all the currents from the memory cells connected to that particular bitline.
(124) VMM array 1400 implements uni-directional tuning for non-volatile memory cells in memory array 1403. That is, each non-volatile memory cell is erased and then partially programmed until the desired charge on the floating gate is reached. This can be performed, for example, using the precision programming techniques described below. If too much charge is placed on the floating gate (such that the wrong value is stored in the cell), the cell must be erased and the sequence of partial programming operations must start over. As shown, two rows sharing the same erase gate (such as EG0 or EG1) need to be erased together (which is known as a page erase), and thereafter, each cell is partially programmed until the desired charge on the floating gate is reached.
(125) Table No. 7 depicts operating voltages for VMM array 1400. The columns in the table indicate the voltages placed on word lines for selected cells, word lines for unselected cells, bit lines for selected cells, bit lines for unselected cells, control gates for selected cells, control gates for unselected cells in the same sector as the selected cells, control gates for unselected cells in a different sector than the selected cells, erase gates for selected cells, erase gates for unselected cells, source lines for selected cells, and source lines for unselected cells. The rows indicate the operations of read, erase, and program.
(126) TABLE-US-00007 TABLE NO. 7 Operation of VMM Array 1400 of FIG. 14 CG - unsel same WL WL - unsel BL BL - unsel CG sector CG - unsel EG EG - unsel SL SL - unsel Read 0.5-2 V −0.5 V/0 V 0.1-2 V 0 V/FLT 0-2.6 V 0-2.6 V 0-2.6 V 0-2.6 V 0-2.6 V 0 V 0 V (Ineuron) Erase 0 V 0 V 0 V 0 V 0 V 0-2.6 V 0-2.6 V 5-12 V 0-2.6 V 0 V 0 V Program 0.7-1 V −0.5 V/0 V 0.1-1 uA Vinh (1-2 V) 4-11 V 0-2.6 V 0-2.6 V 4.5-5 V 0-2.6 V 4.5-5 V 0-1 V
(127)
(128) Table No. 8 depicts operating voltages for VMM array 1500. The columns in the table indicate the voltages placed on word lines for selected cells, word lines for unselected cells, bit lines for selected cells, bit lines for unselected cells, control gates for selected cells, control gates for unselected cells in the same sector as the selected cells, control gates for unselected cells in a different sector than the selected cells, erase gates for selected cells, erase gates for unselected cells, source lines for selected cells, and source lines for unselected cells. The rows indicate the operations of read, erase, and program.
(129) TABLE-US-00008 TABLE NO. 8 Operation of VMM Array 1500 of FIG. 15 CG -unsel same WL WL -unsel BL BL -unsel CG sector CG -unsel EG EG -unsel SL SL -unsel Read 1.0-2 V −0.5 V/0 V 0.6-2 V 0 V/FLT 0-2.6 V 0-2.6 V 0-2.6 V 0-2.6 V 0-2.6 V 0 V 0 V/FLT (Ineuron) Erase 0 V 0 V 0 V 0 V 0 V 4-9 V 0-2.6 V 5-12 V 0-2.6 V 0 V 0 V Program 0.7-1 V −0.5 V/0 V 0.1-1 uA Vinh (1-2 V) 4-11 V 0-2.6 V 0-2.6 V 4.5-5 V 0-2.6 V 4.5-5 V 0-1 V
(130)
(131)
(132)
(133)
(134)
(135)
(136)
(137)
(138)
(139) Long Short-Term Memory
(140) The prior art includes a concept known as long short-term memory (LSTM). LSTMs often are used in artificial neural networks. LSTM allows an artificial neural network to remember information over predetermined arbitrary time intervals and to use that information in subsequent operations. A conventional LSTM comprises 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 that the information is remembered in the LSTM. VMMs are particularly useful in LSTMs.
(141)
(142)
(143) LSTM cell 2600 comprises sigmoid function devices 2601, 2602, and 2603, each of which applies a number between 0 and 1 to control how much of each component in the input vector is allowed through to the output vector. LSTM cell 2600 also comprises tanh devices 2604 and 2605 to apply a hyperbolic tangent function to an input vector, multiplier devices 2606, 2607, and 2608 to multiply two vectors together, and addition device 2609 to add two vectors together. Output vector h(t) can be provided to the next LSTM cell in the system, or it can be accessed for other purposes.
(144)
(145) An alternative to LSTM cell 2700 (and another example of an implementation of LSTM cell 2600) is shown in
(146) Whereas LSTM cell 2700 contains multiple sets of VMM arrays 2701 and respective activation function blocks 2702, LSTM cell 2800 contains only one set of VMM arrays 2801 and activation function block 2802, which are used to represent multiple layers in the embodiment of LSTM cell 2800. LSTM cell 2800 will require less space than LSTM 2700, as LSTM cell 2800 will require ¼ as much space for VMMs and activation function blocks compared to LSTM cell 2700.
(147) It can be further appreciated that LSTM units will typically comprise multiple VMM arrays, each of which requires functionality provided by certain circuit blocks outside of the VMM arrays, such as a summer and activation circuit block and high voltage generation blocks. Providing separate circuit blocks for each VMM array would require a significant amount of space within the semiconductor device and would be somewhat inefficient. The embodiments described below therefore attempt to minimize the circuitry required outside of the VMM arrays themselves.
(148) Gated Recurrent Units
(149) An analog VMM implementation can be utilized for a GRU (gated recurrent unit). GRUs are a gating mechanism in recurrent artificial neural networks. GRUs are similar to LSTMs, except that GRU cells generally contain fewer components than an LSTM cell.
(150)
(151)
(152)
(153) An alternative to GRU cell 3100 (and another example of an implementation of GRU cell 3000) is shown in
(154) Whereas GRU cell 3100 contains multiple sets of VMM arrays 3101 and activation function blocks 3102, GRU cell 3200 contains only one set of VMM arrays 3201 and activation function block 3202, which are used to represent multiple layers in the embodiment of GRU cell 3200. GRU cell 3200 will require less space than GRU cell 3100, as GRU cell 3200 will require ⅓ as much space for VMMs and activation function blocks compared to GRU cell 3100.
(155) It can be further appreciated that systems utilizing GRUs will typically comprise multiple VMM arrays, each of which requires functionality provided by certain circuit blocks outside of the VMM arrays, such as a summer and activation circuit block and high voltage generation blocks. Providing separate circuit blocks for each VMM array would require a significant amount of space within the semiconductor device and would be somewhat inefficient. The embodiments described below therefore attempt to minimize the circuitry required outside of the VMM arrays themselves.
(156) The input to the VMM arrays can be an analog level, a binary level, timing pulses, or digital bits and the output can be an analog level, a binary level, timing pulses, or digital bits (in this case an output ADC is needed to convert output analog level current or voltage into digital bits).
(157) For each memory cell in a VMM array, each weight w can be implemented by a single memory cell or by a differential cell or by two blend memory cells (average of 2 or more cells). In the differential cell case, two memory cells are needed to implement a weight w as a differential weight (w=w+−w−). In the two blend memory cells, two memory cells are needed to implement a weight w as an average of two cells.
(158) Embodiments for Precise Programming of Cells in a VMM
(159) Embodiments for precisely programming memory cells within a VMM by incrementing or decrementing the programming voltages applied to different terminals of the memory cells will now be described.
(160)
(161) A coarse programming method (to get the cell much closer to the target, for example 2×-100× the target) is then performed on the selected cells (step 3305), followed by a precision programming method on the selected cells (step 3306) to program the precise value desired for each selected cell.
(162)
(163) Different terminals of the memory cell can be used for coarse programming method 3305 and precision programming method 3306. That is, during coarse programming method 3305, the voltage applied to one of the terminals of the memory cell (which can be referred to as a coarse programming terminal) will be altered until the desired voltage level is achieved within floating gate 20, and during precision programming method 3306, the voltage applied to one of the terminals of the memory cell (which can be referred to as a precision programming terminal) will be altered until the desired level is achieved. The various combinations of terminals that can be used as the coarse programming terminal and the precision programming terminal are shown in Table 9:
(164) TABLE-US-00009 TABLE NO. 9 Memory Cell Terminals Used for Coarse and Precision Programming Methods Programming Coarse Programming Precision Programming Applicable Memory Sequence Name Terminal Terminal Cell Types CG-CG Control Gate 28 (CG) Control Gate 28 (CG) 310, 410, 510, 710 CG-EG Control Gate 28 (CG) Erase Gate 30 (EG) 410 CG-EG Control Gate 28 (CG) Control Gate 28 (CG) 410 and Erase Gate 30 (EG) or Source Line 14 (SL) EG-EG Erase Gate 30 (EG) Erase Gate 30 (EG) 410, 610 EG-CG Erase Gate 30 (EG) Control Gate 28 (CG) 410 CG-SL Control Gate 28 (CG) Source Line 14 (SL) 310, 410, 510, 710 SL-CG Source Line 14 (SL) Control Gate 28 (CG) 310, 410, 510, 710 SL-EG Source Line 14 (SL) Erase Gate 30 (EG) 410, 610
Other combinations of terminals are possible for the coarse programming and precision programming steps.
(165)
(166) Examples of cell values, desired current values, and coarse target current values are depicted in Tables 10 and 11 for the simple example of N=8 and M=4:
(167) TABLE-US-00010 TABLE NO. 10 Example of N Desired Current Values for N = 8 Value to be Stored in Selected Cell Desired Current Value (I.sub.D) 000 0.5 nA 001 .sup. 1 nA 010 1.5 nA 011 .sup. 2 nA 100 2.5 nA 101 .sup. 3 nA 110 3.5 nA 111 .sup. 4 nA
(168) TABLE-US-00011 TABLE NO. 11 Example of M Target Current Values for M = 4 Coarse Target Current Value (I.sub.CT) Associated Cell Values 4 nA + I.sub.CTOFFSET1 000, 001 8 nA + I.sub.CTOFFSET2 010, 011 12 nA + I.sub.CTOFFSET3 100, 101 16 nA + I.sub.CTOFFSET4 110, 111
The offset values I.sub.CTOFFSETx are used to prevent overshooting the desired current value during coarse tuning.
(169) Once the coarse target current value I.sub.CT is selected, in step 3402 the selected cell is programmed by applying an initial voltage v.sub.0 to the coarse programming terminal of the selected cell according to one of the sequences listed in Table 9, above. (The value of initial voltage v.sub.0 and appropriate coarse programming terminal, optionally can be determined from a voltage look up table that stores v.sub.0 vs. coarse target current value I.sub.CT):
(170) TABLE-US-00012 TABLE NO. 12 Initial Voltage v.sub.0 Applied to Coarse Programming Terminal Initial Voltage v.sub.o Programming Coarse Programming Applied to Coarse Sequence Name Terminal Programming Terminal CG-CG Control Gate 28 (CG) 3-6 V CG-EG Control Gate 28 (CG) 3-6 V EG-EG Erase Gate 30 (EG) 0-3 V CG-SL Control Gate 28 (CG) 3-6 V SL-CG Source Line 14 (SL) 3-4 V SL-EG Source Line 14 (SL) 3-4 V
(171) Next, in step 3403, the selected cell is programmed by applying the voltage v.sub.i=v.sub.i-1+v.sub.increment, to the coarse programming terminal, where i starts at 1 and increments each time this step is repeated, and where v.sub.increment is an increment of coarse voltage that will cause a degree of programming that is appropriate for the granularity of change desired. Thus, the first time step 3403 is performed, i=1, and v.sub.1 will be v.sub.0+v.sub.increment. Then a verify operation occurs (step 3404), wherein a read operation is performed on the selected cell and the current drawn through the selected cell (I.sub.cell) is measured. If I.sub.cell is less than or equal to I.sub.CT (which here is a first threshold value), then search and execute method 3400 is complete and precision programming method 3306 can begin. If I.sub.cell is not less than or equal to I.sub.CT, then step 3403 is repeated, and i is incremented.
(172) Thus, at the point when coarse programming method 3305 ends and precision programming method 3306 begins, the voltage v.sub.i will be the last voltage applied to the coarse programming terminal to program the selected cell, and the selected cell will be storing a value associated with the coarse target current value I.sub.CT, particularly less than or equal to I.sub.CT. The goal of precision programming method 3306 is to program the selected cell to the point where during a read operation it draws current I.sub.D (plus or minus an acceptable margin of deviation, such as +/−30% or less), which is the desired current value that is associated with the value that is intended to be stored in the selected cell.
(173)
(174) Under a first approach, increasing voltages are applied in progression to the coarse programming terminal and/or the precision programming terminal to further program the selected memory cell. The starting point is v.sub.i, which is the last voltage applied during coarse programming method 3305. An increment of v.sub.p1 is added to v.sub.i and the voltage v.sub.i+v.sub.p1 is then used to program the selected cell (indicated by the second pulse from the left in progression 3501). v.sub.p1 is an increment that is smaller than v.sub.increment (the voltage increment used during coarse programming method 3305). After each programming voltage is applied to programming terminal, a verify step (similar to step 3404) is performed, where a determination is made if Icell is less than or equal to I.sub.PT1 (which is the first precision target current value and here is a second threshold value), where I.sub.PT1=I.sub.D+I.sub.PT1OFFSET, where I.sub.PT1OFFSET is an offset valued added to prevent program overshoot. If it is not, then another increment v.sub.p1 is added to the previously-applied programming voltage, and the process is repeated. This repeats until the point where I.sub.cell is less than or equal to I.sub.PT1, at which point this portion of the programming sequence stops. Optionally, if I.sub.PT1 is equal to I.sub.D, or almost equal to I.sub.D with sufficient allowed precision, then the selected memory cell has been successfully programmed.
(175) If I.sub.PT1 is not close enough to I.sub.D, then further programming of a smaller granularity can occur. Here, progression 3502 is now used. The starting point for progression 3502 is the last voltage used for programming under progression 3501. An increment of V.sub.p2 (which is smaller than v.sub.p1) is added to that voltage, and the combined voltage is applied to the precision programming terminal to program the selected memory cell. After each programming voltage is applied, a verify step (similar to step 3404) is performed, where a determination is made if I.sub.cell is less than or equal to I.sub.PT2 (which is the second precision target current value and here is a third threshold value), where I.sub.PT2=I.sub.D+I.sub.PT2OFFSET, I.sub.PT2OFFSET is an offset value added to prevent program overshoot. If it is not, then another increment V.sub.p2 is added to the previously-applied programming voltage, and the process is repeated. This repeats until the point where I.sub.cell is less than or equal to I.sub.PT2, at which point this portion of the programming sequence stops. Here, it is assumed that I.sub.PT2 is equal to I.sub.D or close enough to I.sub.D that the programming can stop, since the target value has been achieved with sufficient allowed precision. One of ordinary skill in the art can appreciate that additional progressions can be applied with smaller and smaller programming increments used. For example, in
(176) A second approach is shown in progression 3503. Here, instead of increasing the voltage applied during the programming of the selected memory cell, the same voltage (such as V.sub.i, or V.sub.i+V.sub.p1+V.sub.p1, or V.sub.i+V.sub.p2+V.sub.p2) is applied for durations of increasing period. Instead of adding an incremental voltage, such as v.sub.p1 in progression 3501 and v.sub.p2 in progression 3502, an additional increment of time t.sub.p1 is added to the programming pulse such that each applied pulse is longer than the previously-applied pulse by t.sub.p1. After each programming pulse is applied to the precision programming terminal, the same verify step is performed as described previously for progression 3501. Optionally, additional progressions can be applied where the additional increment of time added to the programming pulse is of a smaller duration than the previous progression used.
(177) Optionally, additional program cycle progressions can be applied where the programming pulse is of same duration as the previous program cycle progression used. Although only one temporal progression is shown, one of ordinary skill in the art will appreciate that any number of different temporal progressions can be applied. That is, instead of altering the voltage magnitude used during programming or altering the period of a voltage pulse used during programming, the system instead can alter the number of programming cycles used.
(178)
(179) Additional detail will now be provided for second and third embodiments of coarse programming method 3305.
(180)
(181) A new program voltage, v.sub.i, is determined. The first time this step is performed, i=1, and program voltage v.sub.1 is determined based on the stored sub-threshold slope value and a current target and offset value using a sub-threshold equation, such as the following:
vi=v.sub.i-1+v.sub.increment, v.sub.increment is proportional to slope of Vg
Vg=n*Vt*log [Ids/wa*Io]
Here, wa is w of a memory cell, Ids is the cell current, Io is the cell current when Vg=Vth, Vt is thermal voltage, and utilizing g from 1 to 2, where V1 is determined at current IR1 and V2 is determined at current IR2.
slope=(V1−V2)/(LOG(IR1)−LOG(IR2))
where v.sub.increment=alpha*slope*(LOG (IR1)−LOG (I.sub.CT)),
where I.sub.CT is the target current and alpha is a pre-determined constant <1 (programming offset value) to prevent overshoot, e.g., 0.9.
(182) If the stored slope value is relatively steep, then a relatively small current offset value can be used. If the stored slope value is relatively flat, then a relatively high current offset value can be used. Thus, determining the slope information will allow for a current offset value to be selected that is customized for the particular cell in question. This will generally make the programming process shorter. When step 3704 is repeated, i is incremented, and v.sub.i=v.sub.i-1+v.sub.increment. The cell is then programmed by applying v.sub.i to the coarse adaptive programming terminal. v.sub.increment can also be determined from a lookup table storing values of v.sub.increment. vs. target current value.
(183) Next, a verify operation occurs, wherein a read operation is performed on the selected cell and the current drawn through the selected cell (I.sub.cell) is measured (step 3705). If I.sub.cell is less than or equal to I.sub.CT (which here is a coarse target threshold value), where I.sub.CT=I.sub.D+I.sub.CTOFFSET, where I.sub.CTOFFSET is an offset value added to prevent program overshoot, then adaptive calibration method 3700 is complete and precision programming method 3306 can begin. If I.sub.cell is not less than or equal to I.sub.CT, then steps 3704-3705 (where a new slope measurement is measured with new data points) or steps 3703-3705 (where the same slope used previously is re-used) are repeated, and i is incremented.
(184)
(185)
(186) TABLE-US-00013 TABLE NO. 13 Initial Voltage v.sub.0 Applied to Coarse Adaptive Programming Terminal During Adaptive Calibration Method 3900 Initial Voltage v.sub.o Applied Programming Coarse Adaptive to Coarse Sequence Programming Programming Name Terminal Terminal V1 V2 CG-CG Control Gate 28 (CG) ~5-7 V ~1.3 ~1.5 V CG-EG Control Gate 28 (CG) ~5-7 V ~1.3 ~1.5 V EG-EG Erase Gate 30 (EG) ~0-2 V ~1.3 ~1.5 V CG-SL Control Gate 28 (CG) ~5-7 V ~1.3 ~1.5 V SL-CG Source Line 14 (SL) ~3-3.5 V.sup. ~1.3 ~1.5 V SL-EG Source Line 14 (SL) ~3-3.5 V.sup. ~1.3 ~1.5 V
(187) In step 3903 an I-V slope parameter is created which is used in predicting the next programming voltage. A first voltage, V1, is applied to the control gate or erase gate of the selected cell, and the resulting cell current, IR.sub.1, is measured. Then a second voltage, V2, is applied to the control gate or erase gate of the selected cell, and the resulting cell current, IR2, is measured. A slope is determined based on those measurements and stored, for example as according to the equation in sub-threshold region (cell operating in sub-threshold):
slope=(V1−V2)/(LOG(IR1)−LOG(IR2))
(step 3903). Examples of values for V1 and V2 are shown in Table No. 13, above.
(188) Determining the I-V slope information allows for a v.sub.increment value to be selected that is customized for the particular cell in question. This generally will make the programming process shorter.
(189) Each time step 3904 is performed, i is incremented, with an initial value of 0, to determine a desired programming voltage, v.sub.i, based on the stored slope value and a current target and offset value using an equation such as the following:
v.sub.i=v.sub.i-1+v.sub.increment,
where v.sub.increment=alpha*slope*(LOG(IR.sub.1)−LOG(I.sub.CT)),
where I.sub.CT is the target current and alpha is a pre-determined constant <1 (programming offset value) to prevent overshoot, e.g., 0.9.
(190) The selected cell is then programmed using v.sub.i. (step 3905)
(191) Next, a verify operation occurs, wherein a read operation is performed on the selected cell and the current drawn through the selected cell (I.sub.cell) is measured (step 3906). If I.sub.cell is less than or equal to I.sub.CT (which here is a coarse target threshold value), where I.sub.CT=I.sub.D+I.sub.CTOFFSET, where I.sub.CTOFFSET is an offset value added to prevent program overshoot, then the process proceeds to the step 3907. If not, then the process returns to step 3903 (new slope measurement) or 3904 (previous slope re-used) and i is incremented.
(192) In step 3907, I.sub.cell is compared against a threshold value, I.sub.CT2, that is smaller than I.sub.CT. The purpose of this is to see if an overshoot has occurred. That is, although the goal is for I.sub.cell to be below I.sub.CT, if it falls too far below I.sub.CT, then an overshoot has occurred and the stored value may actually correspond to the wrong value. If I.sub.cell is not less than or equal to I.sub.CT2, then no overshoot has occurred, and adaptive calibration method 3900 has completed, as which point the process progresses to precision programming method 3306. If I.sub.cell is less than or equal to I.sub.CT2, then an overshoot has occurred. The selected cell is then erased (step 3908), and the programming process starts over at step 3902. Optionally, if step 3908 is performed more than a predetermined number of times, the selected cell can be deemed a bad cell that should not be used, and an error signal is output, or a flag set, identifying the cell.
(193) The precision program method 3306 may consist of multiple verify and program cycles, in which the program voltage is incremented by a constant fine voltage with a fixed pulse width or in which the program voltage is fixed and the program pulse width is varied for each additional programming pulse.
(194) Optionally, step 3906 of determining if the current through the selected non-volatile memory cell during a read or verify operation is less than or equal to the first threshold current value can be performed by applying fixed biases to terminals of the non-volatile memory cell, measuring and digitizing the current drawn by the selected non-volatile memory cell to generate digital output bits, and comparing the digital output bits to digital bits representing the first threshold current, I.sub.CT.
(195) Optionally, step 3907 of determining if the current through the selected non-volatile memory cell during a read or verify operation is less than or equal to the second threshold current value can be performed by applying fixed biases to terminals of the non-volatile memory cell, measuring and digitizing the current drawn by the selected non-volatile memory cell to generate digital output bits, and comparing the digital output bits to digital bits representing the second threshold current, I.sub.CT2.
(196) Optionally, each of step 3906, 3907 of determining if the current through the selected non-volatile memory cell during a read or verify operation is less than or equal to the first, or second threshold current value, respectively, can be performed by applying an input to a terminal of the non-volatile memory cell, modulating the current drawn by the selected non-volatile memory cell with an output pulse to generate a modulated output, digitizing the modulated output to generate digital output bits, and comparing the digital output bits to digital bits representing the first or second threshold current value, respectively.
(197) Measuring cell current for the purpose of verifying or reading the current can be done by averaging over multiple times, e.g., 8-32 times, to reduce the impact of noise.
(198)
(199) TABLE-US-00014 TABLE NO. 14 Initial Voltage v.sub.0 Applied to Memory Cell Terminal During Absolute Calibration Method 4000 Initial Voltage v.sub.o Programming Coarse Programming Applied to Coarse Sequence Name Terminal Programming Terminal CG-CG Control Gate 28 (CG) ~5-7 V CG-EG Control Gate 28 (CG) ~5-7 V EG-EG Erase Gate 30 (EG) ~0-2 V CG-SL Control Gate 28 (CG) ~5-7 V SL-CG Source Line 14 (SL) ~3-3.5 V.sup. SL-EG Source Line 14 (SL) ~3-3.5 V.sup.
(200) The voltage vTx on the coarse programming terminal is measured at a current value Itarget driven through the cell as described above in relation to
(201) The cell is then programmed using v.sub.i (step 4005). When i=1, the voltage v.sub.1 from step 4004 is used. When i>=2, the voltage v.sub.i=v.sub.i-1+v.sub.increment is used. v.sub.increment can be determined from a lookup table storing values of v.sub.increment. vs. target current value. Next, a verify operation occurs, wherein a read operation is performed on the selected cell and the current drawn through the selected cell (I.sub.cell) is compared to I.sub.CT (step 4006). If I.sub.cell is less than or equal to I.sub.CT (which here is a threshold value), then absolute calibration method 4000 is complete and precision programming method 3306 can begin. If I.sub.cell is not less than or equal to I.sub.CT, then steps 4005-4006 are repeated, and i is incremented.
(202)
(203)
(204)
(205) In another embodiment, the voltage applied to the control gate terminal is incremented and the voltage applied to the erase gate terminal is also incremented.
(206) In another embodiment, the voltage applied to the control gate terminal is incremented and the voltage applied to the erase gate terminal is decremented. This is illustrated in Table 15:
(207) TABLE-US-00015 TABLE 15 Control Gate Terminal Increment and Erase Gate Terminal Decrement VCG VEG CG-FG delta VFG mV EG-FG delta VFG mV increment decrement coupling due to VCG coupling due to VEG total delta mV mV ratio increment ratio decrement VFG mV 10 −10 40% 4 15% −1.5 2.5 10 −20 40% 4 15% −3 1 20 −50 40% 8 15% −7.5 0.5
(208) For comparison's sake, an example of incrementing only the control gate terminal or incrementing only the erase gate terminal are contained in Table 16:
(209) TABLE-US-00016 TABLE 16 Control Gate Terminal Increment; Erase Gate Terminal Increment veg increment mV CG-FG coupling ratio delta VFG mV Case 1: VCG increment 10 40% 4 Case 2: VEG increment 10 15% 1.5
(210)
(211) In one embodiment, the digital values provided to input function circuit 4401 comprise four bits (DIN3, DIN2, DIN1, and DIN0), or any number of bits, and the digital value represented by those bits corresponds to the number of input pulses to be applied to the control gate during a programming operation. A greater number of pulses will cause a greater value to be stored in the cell, which will cause a larger output current when the cell is read. An example of input bit values and pulse values is shown in Table No. 17:
(212) TABLE-US-00017 TABLE NO. 17 Digital Bit Inputs v. Pulses Generated Input Pulses DIN3 DIN2 DIN1 DIN0 Generated 0 0 0 0 0 0 0 0 1 1 0 0 1 0 2 0 0 1 1 3 0 1 0 0 4 0 1 0 1 5 0 1 1 0 6 0 1 1 1 7 1 0 0 0 8 1 0 0 1 9 1 0 1 0 10 1 0 1 1 11 1 1 0 0 12 1 1 0 1 13 1 1 1 0 14 1 1 1 1 15
(213) In the above example, there are a maximum of 15 pulses for 4 bit input digital. Each pulse is equal to one unit cell value (current), i.e. the precision programmed current. For example, if Icell unit=1 nA, then for DIN[3-0]=0001, Icell=1*1 nA=1 nA; and for DIN[3-0]=1111, Icell=15*1 nA=15 nA.
(214) In another embodiment, the digital bit input uses digital bit position summation to read out the cell or neuron (e.g., precisely programmed value on bitline output) value as shown in Table 18. Here, only 4 pulses or 4 fixed same bias inputs (for example input on wordline or control gate) are needed to evaluate the 4 bit digital value. For example, a first pulse or a first fixed bias is used to evaluate DIN0, a second pulse or a second fixed bias with same value as the first one is used to evaluate DIN1, a third pulse or a third fixed bias with same value as the first one is used to evaluate DIN2, and a fourth pulse or a fourth fixed bias with same value as the first one is used to evaluate DIN3. Then, the results from the four pulses are summed according to bit position with each output result multiplied (scaled) by a multiplier factor that is 2{circumflex over ( )}n, n is the digital bit position as shown in Table 19. The digital bit summation equation realized is the following: Output=2{circumflex over ( )}0*DIN0+2{circumflex over ( )}1*DIN1+2{circumflex over ( )}2*DIN2+2{circumflex over ( )}3*DIN3)*Icell unit, where Icell unit represent the precision programmed current.
(215) For example, if Icell unit=1 nA, then for DIN[3-0]=0001, Icell total=0+0+0+1*1 nA=1 nA; and for DIN[3-0]=1111, Icell total=8*1 nA+4*1 nA+2*1 nA+1*1 nA=15 nA.
(216) TABLE-US-00018 TABLE NO. 18 Digital Bit Input Summation 2{circumflex over ( )}3 * 2{circumflex over ( )}2 * 2{circumflex over ( )}1 * 2{circumflex over ( )}0 * DIN3 DIN2 DIN1 DIN0 Total values 0 0 0 0 0 0 0 0 1 1 0 0 2 0 2 0 0 2 1 3 0 4 0 0 4 0 4 0 1 5 0 4 2 0 6 0 4 2 1 7 8 0 0 0 8 8 0 0 1 9 8 0 2 0 10 8 0 2 1 11 8 4 0 0 12 8 4 0 1 13 8 4 2 0 14 8 4 2 1 15
(217) TABLE-US-00019 TABLE 19 Digital input bit Dn summation with 2{circumflex over ( )}n output multiplication factor DIN3 DIN2 DIN1 DIN0 Output X factor Y X1 Y X2 Y X4 Y X8
(218) Another embodiment with a hybrid input with multiple digital input pulse ranges and input digital range summations is shown in Table 20 for an exemplary 4-bit digital input. In this embodiment, DINn-0 can be divided into m different groups, where each group is evaluated and the output is scaled by a multiplication factor by the group binary position. As example, for 4-bit DIN3-0, the groups can be DIN3-2 and DIN1-0, where the output for DIN1-0 is scaled by one (×1) and the output for DIN3-2 is scaled by 4 (×4).
(219) TABLE-US-00020 TABLE 20 Hybrid input-output summation with multiple input ranges input pulses (or equivalent Output X DIN3 DIN2 DIN1 DIN0 pulsewidth) factor 0 0 0 X1 0 1 1 X1 1 0 2 X1 1 1 3 X2 0 0 0 X4 0 1 1 X4 1 0 2 X4 1 1 3 X4
(220) Another embodiment combines a hybrid input range with a hybrid supercell. A hybrid super cells includes multiple physical x-bit cells to implement a logical n-bit cell with the x-cell output scaled by the 2{circumflex over ( )}n binary position. For example, to implement an 8-bit logical cell, two 4-bit cells (cell1, cell0) are used. The output for cell0 is scaled by one (×1) and the output for cell1 is scaled by four (X, 2{circumflex over ( )}2). Other combinations of physical x-cells to implement n-bit logical cell are possible such as two 2-bit physical cell and one 4-bit physical cell to implement an 8-bit logical cell.
(221)
(222)
(223)
(224)
(225)
(226)
(227)
(228)
(229)
(230) A Modified Binary Search such as a cyclic (algorithmic) ADC can be used for the cell tuning (e.g., programming) verification or the output neuron conversion. A Modified Binary Search such as a switched cap (SC) charge re-distribution ADC can be used for the cell tuning (e.g., programming) verification or the output neuron conversion.
(231)
(232)
(233) Alternatively, ramp voltage Vreframp is a continuous ramp voltage 5255 as shown in graph 5250 of
(234) Alternatively, a multi-ramp embodiment is shown in
(235)
(236) In another embodiment, a hybrid ADC can be used. For example, for a 9-bit ADC, the first 4 bits can be generated by a SAR ADC and the remaining 5 bits can be generated using a slope ADC or a ramp ADC.
(237) Programming and Verifying Multiple Physical Cells as a Single Logical Multi-Bit Cell
(238) The programming and verifying devices and methods described above can operate concurrently upon multiple physical cells as a logical multi-bit cell.
(239)
(240)
(241) Method 5500 can be performed on more than one subset of the i physical cells 5401-1, . . . 5401-i to achieve the desired overall level for logical multi-bit cell 5400.
(242) For example, if i=4, then there will be four cells, 5401-1, 5401-2, 5401-3, and 5401-4. If one assumes that each cell can hold one of 8 different levels, then logical multi-bit cell 5400 can hold one of 32 different levels. If the desired programming value is L27, then that level (which corresponds to a desired read current) can be achieved in any number of different ways.
(243) For example, method 5500 can be performed on cells 5401-1, 5401-2, and 5401-3 until those cells collectively hold L23 (the 24th level), and then method 5500 can be performed on cell 5401-4 to program that cell to its fourth level so that the logical multibit cell 5400 achieves L27 (the 28th level).
(244) As another example, method 5500 can be performed on cells 5401-1, 5401-2, 5401-3, and 5401-4 until those cells collectively hold L25 (the 26th level), and then method 5500 can be performed only on cell 5401-4 until it stores a value that causes the entirety of logical multi-bit cell 5400 to achieve L27 (the 28th level).
(245) Other approaches are possible, and method 5500 can be performed on different subsets of the i physical cells until the desired level is achieved.
(246) In another embodiment, in the situation where the i physical cells have a non-uniform diffusion width, a coarse programming step 3305 can be performed on the j1 physical cells with the wider transistor width until the j1 physical cells collectively achieve a coarse current target, and then a precision programming step 3306 can be performed on the j2 physical cells with the smallest transistor width until the j1+j2 physical cells collectively achieve a precision current target.
(247) It should be noted that, as used herein, the terms “over” and “on” both inclusively include “directly on” (no intermediate materials, elements or space disposed therebetween) and “indirectly on” (intermediate materials, elements or space disposed therebetween). Likewise, the term “adjacent” includes “directly adjacent” (no intermediate materials, elements or space disposed therebetween) and “indirectly adjacent” (intermediate materials, elements or space disposed there between), “mounted to” includes “directly mounted to” (no intermediate materials, elements or space disposed there between) and “indirectly mounted to” (intermediate materials, elements or spaced disposed there between), and “electrically coupled” includes “directly electrically coupled to” (no intermediate materials or elements there between that electrically connect the elements together) and “indirectly electrically coupled to” (intermediate materials or elements there between that electrically connect the elements together). For example, forming an element “over a substrate” can include forming the element directly on the substrate with no intermediate materials/elements therebetween, as well as forming the element indirectly on the substrate with one or more intermediate materials/elements there between.