Abstract: Numerous embodiments are disclosed for programmable output blocks for use with a VMM array within an artificial neural network. In one embodiment, the gain of an output block can be configured by a configuration signal. In another embodiment, the resolution of an ADC in the output block can be configured by a configuration signal.

Abstract: Numerous embodiments are disclosed for providing temperature compensation in a an analog memory array. The analog memory array optionally is a vector-by-matrix multiplier in an analog neuromorphic memory system used in a deep learning neural network. One embodiment comprises measuring an operating temperature within a memory array and applying, by a temperature compensation block, a bias voltage to a terminal of a memory cell in the array, wherein the bias voltage is a function of the operating temperature.

Abstract: Various algorithms are disclosed for verifying the stored weight in a non-volatile memory cell in a neural network following a multilevel programming operation of the non-volatile memory cell by converting the stored weight into a plurality of digital output bits. Circuity, such as an adjustable reference current source, for implementing the algorithms are disclosed.

Abstract: Memory cells formed on upwardly extending fins of a semiconductor substrate, each including source and drain regions with a channel region therebetween, a floating gate extending along the channel region and wrapping around the fin, a word line gate extending along the channel region and wrapping around the fin, a control gate over the floating gate, and an erase gate over the source region. The control gates are a continuous conductive strip of material. First and second fins are spaced apart by a first distance. Third and fourth fins are spaced apart by a second distance. The second and third fins are spaced apart by a third distance greater than the first and second distances. The continuous strip includes a portion disposed between the second and third fins, but no portion of the continuous strip is disposed between the first and second fins nor between the third and fourth fins.

Abstract: Numerous embodiments for performing tuning of a page or a word of non-volatile memory cells in an analog neural memory are disclosed. High voltage circuits used to generate high voltages applied to terminals of the non-volatile memory cells during the precision tuning process are also disclosed. Programming sequences for the application of the voltages to the terminals to minimize the occurrence of disturbances during tuning are also disclosed.

Abstract: A number of circuits for use in an output block coupled to a non-volatile memory array in a neural network are disclosed. The embodiments include a circuit for converting an output current from a neuron in a neural network into an output voltage, a circuit for converting a voltage received on an input node into an output current, a circuit for summing current received from a plurality of neurons in a neural network, and a circuit for summing current received from a plurality of neurons in a neural network.

Abstract: Numerous embodiments of an improved charge pump design are disclosed for generating the high voltages necessary to perform erase and program operations in non-volatile flash memory devices. In these embodiments, each boost stage in the charge pump is modified to overcome a deficiency in prior art charge pumps whereby voltage actually would decrease in the final boost stage. These modifications include the addition of one or more of a clock doubling circuit, a local self-precharge circuit, a feed-forward precharge circuit, a feed-backward precharge circuit, and a hybrid circuit comprising NMOS and PMOS transistors and diodes.

Abstract: Numerous embodiments of analog neural memory arrays are disclosed. In certain embodiments, each memory cell in the array has an approximately constant source impedance when that cell is being operated. In certain embodiments, power consumption is substantially constant from bit line to bit line within the array when cells are being read. In certain embodiments, weight mapping is performed adaptively for optimal performance in power and noise.

Abstract: Numerous embodiments of analog neural memory arrays are disclosed. In one embodiment, an analog neural memory system comprises an array of non-volatile memory cells, wherein the cells are arranged in rows and columns, the columns arranged in physically adjacent pairs of columns, wherein within each adjacent pair one column in the adjacent pair comprises cells storing W+ values and one column in the adjacent pair comprises cells storing W? values, wherein adjacent cells in the adjacent pair store a differential weight, W, according to the formula W=(W+)?(W?). In another embodiment, an analog neural memory system comprises a first array of non-volatile memory cells storing W+ values and a second array storing W? values.

Abstract: Embodiments for ultra-precise tuning of a selected memory cell are disclosed. The selected memory cell optionally is first programmed using coarse programming and fine programming methods. The selected memory cell then undergoes ultra-precise programming through the programming of an adjacent memory cell. As the adjacent memory cell is programmed, capacitive coupling between the floating gate of the adjacent memory cell and the floating gate of the selected memory cell will cause the voltage of the floating gate of the selected memory cell to increase, but in smaller increments than could be achieved by programming the selected memory cell directly. In this manner, the selected memory cell can be programmed with ultra-precise gradations.

Abstract: Numerous embodiments are disclosed for a high voltage generation algorithm and system for generating high voltages necessary for a particular programming operation in analog neural memory used in a deep learning artificial neural network. Compensation measures can be utilized that compensate for changes in voltage or current as the number of cells being programmed changes.

Abstract: Numerous embodiments of circuitry for a set-while-verify operation and a reset-while verify operation for resistive random access memory cells are disclosed. In one embodiment, a set-while-verify circuit for performing a set operation on a selected RRAM cell in the array applies a combination of voltages or current to a bit line, word line, and source line associated with the selected RRAM cell and stops said applying when the set operation is complete. In another embodiment, a reset-while-verify circuit for performing a reset operation on a selected RRAM cell in the array applies a combination of voltages or current to a bit line, word line, and source line associated with the selected RRAM cell and stops said applying when the reset operation is complete.

Abstract: Numerous embodiments of decoders for use with a vector-by-matrix multiplication (VMM) array in an artificial neural network are disclosed. The decoders include bit line decoders, word line decoders, control gate decoders, source line decoders, and erase gate decoders. In certain embodiments, a high voltage version and a low voltage version of a decoder is used.

Abstract: The present invention relates to a flash memory cell with only four terminals and a high voltage row decoder for operating an array of such flash memory cells. The invention allows for fewer terminals for each flash memory cell compared to the prior art, which results in a simplification of the decoder circuitry and overall die space required per flash memory cells. The invention also provides for the use of high voltages on one or more of the four terminals to allow for read, erase, and programming operations despite the lower number of terminals compared to prior art flash memory cells.

Abstract: Numerous embodiments are disclosed for verifying a weight programmed into a selected non-volatile memory cell in a neural memory. In one embodiment, a circuit for verifying a weight programmed into a selected non-volatile memory cell in a neural memory comprises a converter for converting a target weight into a target current and a comparator for comparing the target current to an output current from the selected non-volatile memory cell during a verify operation. In another embodiment, a circuit for verifying a weight programmed into a selected non-volatile memory cell in a neural memory comprises a digital-to-analog converter for converting a target weight comprising digital bits into a target voltage, a current-to-voltage converter for converting an output current from the selected non-volatile memory cell during a verify operation into an output voltage, and a comparator for comparing the output voltage to the target voltage during a verify operation.

Abstract: An improved method and apparatus for programming advanced nanometer flash memory cells is disclosed. In one embodiment, a programming circuit comprises a switch configured to couple a current source to a capacitor during a first mode and to uncouple the current source from the capacitor during the second mode, wherein during the second mode the capacitor is coupled to the gate of a transistor used to program a memory cell.

Abstract: Numerous embodiments are provided for compensating for drift error in non-volatile memory cells within a VMM array in an analog neuromorphic memory system. For example, in one embodiment, a circuit is provided for compensating for drift error during a read operation, the circuit comprising a data drift monitoring circuit coupled to the array for generating an output indicative of data drift; and a bitline compensation circuit for generating a compensation current in response to the output from the data drift monitoring circuit and injecting the compensation current into one or more bitlines of the array.

Abstract: An improved method and apparatus for programming advanced nanometer flash memory cells is disclosed. In one embodiment, a programming circuit comprises a switch configured to couple a current source to a capacitor during a first mode and to uncouple the current source from the capacitor during the second mode, wherein during the second mode the capacitor is coupled to the gate of a transistor used to program a memory cell.

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.

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.