A semiconductor package includes a semiconductor package includes first, second, third and fourth semiconductor chips sequentially stacked on one another. Each of the first, second, third and fourth semiconductor chips includes a first group of bonding pads and a second group of bonding pads alternately arranged in a first direction and input/output (I/O) circuitry selectively connected to the first group of bonding pads respectively. Each of the first, second and third semiconductor chips includes a first group of through electrodes electrically connected to the first group of bonding pads and a second group of through electrodes electrically connected to the second group of bonding pads.

A semiconductor device includes a memory circuit, a first FIFO, a second FIFO and an input/output circuit. The memory circuit outputs data. The first FIFO receives data from the memory circuit and outputs data synchronously with a first clock signal. The second FIFO receives data output from the first FIFO and outputs data synchronously with the first clock signal. The input/output circuit outputs data output from the second FIFO. The second FIFO is disposed in the vicinity of the input/output circuit than the first FIFO.

A semiconductor memory device includes: first pad transmitting and receiving first timing signal; second pad transmitting and receiving data signal in response to the first timing signal; third pad receiving second timing signal; fourth pad receiving control information in response to the second timing signal; memory cell array; sense amplifier connected to the memory cell array; first register connected to the sense amplifier; second register storing first control information; third register storing second control information; and control circuit executing data-out operation. The first control information is stored in the second register based on an input to the fourth pad in response to the second timing signal consisting of i cycles, and the second control information is stored in the third register based on an input to the fourth pad in response to the second timing signal consisting of j cycles.

A method can include, in an integrated circuit device: at a unidirectional command-address (CA) bus having no more than four parallel inputs, receiving a sequence of no less than three command value portions; latching each command value portion in synchronism with rising edges of a timing clock; determining an input command from the sequence of no less than three command value portions; executing the input command in the integrated circuit device; and on a bi-directional data bus having no more than six data input/outputs (IOs), outputting and inputting sequences of data values in synchronism with rising and falling edges of the timing clock. Corresponding devices and systems are also disclosed.

Techniques to couple a high bandwidth memory device on a silicon substrate and a package substrate are disclosed. Examples include selectively activating input/out (I/O) or command and address (CA) contacts on a bottom side of a logic layer for the high bandwidth device based on a mode of operation. The I/O and CA contacts are for accessing one or more memory devices include in the high bandwidth memory device via one or more data channels.

Apparatuses and methods for providing data from stacked memory are described. The stacked memory may include multiple die. In some examples, a master die may receive data from one or more slave die. The master die may provide data from the master die and the data from the one or more slave die to a plurality of output terminals. Different ones of the output terminals may provide data from a different die of the stacked memory. In some examples, the data may be retrieved from the multiple die concurrently.

A memory device includes a plurality of data input/output (I/O) groups each including data I/O circuits, each data I/O circuit comprising a transistor having a predetermined threshold voltage according to a bulk voltage supplied to a bulk terminal thereof; a control circuit suitable for generating a control signal according to a data I/O mode; and a plurality of voltage supply circuits suitable for independently supplying bulk voltages to the plurality of data I/O groups, and changing, in response to the control signal, a level of a bulk voltage corresponding to data I/O groups unused in the data I/O mode, among the plurality of data I/O groups.

Apparatuses and methods for writing data to a memory array are disclosed. When data is duplicative across multiple data lines, data may be transferred across a single line of a bus rather than driving the duplicative data across all of the data lines. The data from the single data line may be provided to the write amplifiers of the additional data lines to provide the data from all of the data lines to be written to the memory. In some examples, error correction may be performed on data from the single data line rather than all of the data lines.

A memory device includes: first power pins in a first power area and configured to receive a first power voltage; data pins configured to transmit or receive data signals, the data pins being arranged in a first region and in a second region each including the first power area; control pins configured to transmit or receive control signals in the first region and in the second region; second power pins in a second power area between the first region and the second region and configured to receive a second power voltage different from the first power voltage; and ground pins in the second power area and configured to receive a ground voltage.

Methods of performing mixed-signal/analog multiply-accumulate (MAC) operations used for matrix multiplication in fully connected artificial neural networks in integrated circuits (IC) are described in this disclosure having traits such as: (1) inherently fast and efficient for approximate computing due to current-mode signal processing where summation is performed by simply coupling wires, (2) free from noisy and power hungry clocks with asynchronous fully-connected operations, (3) saving on silicon area and power consumption for requiring neither any data-converters nor any memory for intermediate activation signals, (4) reduced dynamic power consumption due to Compute-In-Memory operations, (5) avoiding over-flow conditions along key signals paths and lowering power consumption by training MACs in neural networks in such a manner that the population and or combinations of multi-quadrant activation signals and multi-quadrant weight signals follow a programmable statistical distribution profile, (6) programmable current consumption versus degree of precision/approximate computing, (7) suitable for ‘always-on’ operations and capable of ‘self power-off’, (8) inherently simple arrangement for non-linear activation operations such as Rectified Linear Unit, ReLu, and (9) manufacturable on main-stream, low cost, and lagging edge standard digital CMOS process requiring neither any resistors nor any capacitors.