A memory device includes an active region, a select gate, a control gate, and a blocking layer. The active region includes a bottom portion and a protruding portion protruding from the bottom portion. A source is in the bottom portion and a drain is in the protruding portion. The select gate is above the bottom portion. A top surface of the select gate is lower than a top surface of the protruding portion. The control gate is above the bottom portion. The blocking layer is between the select gate and the control gate.

A neural network computing engine having an array of charge-trap-transistor (CTT) elements which are utilized as analog multipliers with all weight values preprogrammed into each CTT element as a CTT threshold voltage, with multiplicator values received from the neural network inference mode. The CTT elements perform computations of a fully connected (FC) neural network with each CTT element representing a neuron. Row resistors for each row of CTT element sum output currents as partial summation results. Counted pulse generators write weight values under control of a pulse generator controller. A sequential analog fabric (SAF) feeds multiple drain voltages in parallel to the CTT array to enable parallel analog computations of neurons. Partial summation results are read by an analog-to-digital converter (ADC).

A semiconductor device including: a first silicon layer including a first single crystal silicon and a plurality of first transistors; a first metal layer disposed over the first silicon layer; a second metal layer disposed over the first metal layer; a third metal layer disposed over the second metal layer; a second level including a plurality of second transistors, the second level disposed over the third metal layer; a fourth metal layer disposed over the second level; a fifth metal layer disposed over the fourth metal layer, a connection path from the fifth metal layer to the second metal layer, where the connection path includes a via disposed through the second level, where the via has a diameter of less than 450 nm, where the fifth metal layer includes a global power distribution grid, and where a typical thickness of the fifth metal layer is greater than a typical thickness of the second metal layer by at least 50%.

A non-volatile memory (NVM) cell includes a semiconductor wire including a select gate portion and a control gate portion. The NVM cell includes a select transistor formed with the select gate portion and a control transistor formed with the control gate portion. The select transistor includes a gate dielectric layer disposed around the select gate portion and a select gate electrode disposed on the gate dielectric layer. The control transistor includes a stacked dielectric layer disposed around the control gate portion, a gate dielectric layer disposed on the stacked dielectric layer and a control gate electrode disposed on the gate dielectric layer. The stacked dielectric layer includes a first silicon oxide layer disposed on the control gate portion, a charge trapping layer disposed on the first silicon oxide, and a second silicon oxide layer disposed on the charge trapping layer.

A method of manufacturing a semiconductor device includes: forming a conductive pad region over a substrate; depositing a dielectric layer over the conductive pad region; forming a first passivation layer over the dielectric layer; etching the first passivation layer through the dielectric layer, thereby exposing a first area of the conductive pad region; forming a second passivation layer over the first area of the conductive pad region; and removing portions of the second passivation layer to expose a second area of the conductive pad region.

A memory device includes a memory cell, a writing transistor, and a reading transistor. The memory cell includes a semiconductor substrate, a tunneling layer, a storage layer, a first electrode, a second electrode, and a third electrode. The tunneling layer is over the semiconductor substrate. The storage layer is on the tunneling layer. The first electrode is on the storage layer. The second electrode is on the tunneling layer. The storage layer has a sidewall facing the second electrode. The third electrode is spaced apart from the second electrode. The writing transistor is electrically connected to the first electrode of the memory cell. The reading transistor is electrically connected to the second electrode of the memory cell.

Numerous examples of an input function circuit block and an output neuron circuit block coupled to a vector-by-matrix multiplication (VMM) array in an artificial neural network are disclosed. In one example, an artificial neural network comprises a vector-by-matrix multiplication array comprising a plurality of non-volatile memory cells organized into rows and columns; an input function circuit block to receive digital input signals, convert the digital input signals into analog signals, and apply the analog signals to control gate terminals of non-volatile memory cells in one or more rows of the array during a programming operation; and an output neuron circuit block to receive analog currents from the columns of the array during a read operation and generate an output signal.

A three-dimensional semiconductor memory device includes a substrate, an electrode structure including gate electrodes sequentially stacked on the substrate, a source structure between the electrode structure and the substrate, vertical semiconductor patterns passing through the electrode structure and the source structure, a data storage pattern between each of the vertical semiconductor patterns and the electrode structure, and a common source pattern between the source structure and the substrate. The common source pattern has a lower resistivity than the source structure and is connected to the vertical semiconductor patterns through the source structure.

Some embodiments include apparatuses and methods of forming the apparatuses. One of the apparatuses includes a memory cell included in a memory cell string; the memory cell including charge storage structure and channel structure separated from the charge storage structure by a dielectric structure; a first control gate associated with the memory cell and located on a first side of the charge storage structure and a first side of the channel structure; and a second control gate associated with the memory cell and electrically separated from the first control gate, the second control gate located on a second side of the charge storage structure and a second side of the channel structure.

Numerous embodiments of a precision tuning 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.