Abstract: The present embodiment provides a physical layer (PHY) between a high-speed memory and a memory controller, including: an analog physical layer including a read data strobe (RDQS) delay unit that delays an RDQS signal of the high-speed memory and outputs the delayed RDQS signal and a data (DQ) delay unit that delays DQ signals and outputs the delayed DQ signals; and a digital physical layer including an asynchronous first-in first-out (FIFO) that samples the DQ signal with the RDQS signal and outputs the DQ signal, a DQ arrangement block that rearranges the DQ signal, and a validity signal forming unit that forms a validity signal indicating validity of data from data output from the asynchronous FIFO, in which the analog physical layer receives a clock having the same frequency as the high-speed memory and operates, and the asynchronous FIFO and the validity signal forming unit receive a clock having a lower frequency than the clock provided to the high-speed memory and operate.

Abstract: The present embodiment relates to a computing apparatus for computing an interpolated non-linear activation function for an input. The computing apparatus includes a plurality of unit processing elements (PEs), and each unit PE includes: a multiplier that multiplies the input and an output of an accumulator, an adder that adds the output of the multiplier and a coefficient of the interpolated non-linear activation function; and an accumulator that accumulates and outputs the output of the adder.

Abstract: A method for ZQ calibration for a data transmission driving circuit of each memory die in a memory chip package in which memory dies are stacked, includes generating a reference current through a reference resistor connected between a power terminal supplying a power voltage of the data transmission driving circuit and a ground terminal and a first transistor that is diode-connected; supplying first currents corresponding to the reference currents to a pull-up driver of each memory die; performing ZQ calibration of a pull-up driver of a corresponding memory die by comparing a first voltage formed by each first current with a reference voltage formed by the reference current in each of the plurality of memory dies; and performing ZQ calibration of a pull-down driver of the corresponding memory die based on an output impedance of the ZQ calibrated pull-up driver in each of the memory dies.

Abstract: An apparatus for receiving a strobe signal may include an amplifier for amplifying a strobe signal input thereto, an offset generator for controlling the setting of a threshold for detecting a preamble signal by generating an offset for the amplifier, and a preamble detector for detecting a first preamble signal occurring at a point at which the amplified strobe signal is equal to or greater than the threshold and turning off the offset generator when the first preamble signal is detected.

Abstract: Provided is a memory interface device. A memory interface device, comprising: a DQS input buffer configured to receive input data strobe signals and output a first intermediate data strobe signal, the DQS input buffer providing a static offset; an offset control circuit configured to receive the first intermediate data strobe signal and output a second intermediate data strobe signal; and a duty adjustment buffer configured to receive the second intermediate data strobe signal and output a clean data strobe signal, wherein the offset control circuit provides a dynamic offset using the clean data strobe signal.

Abstract: Disclosed herein are a method for supporting cache coherency based on virtual addresses for an artificial intelligence processor having large on-chip memory and an apparatus for the same. The method for supporting cache coherency according to an embodiment of the present disclosure includes, by an artificial intelligence processor including multiple processor cores and multiple caches, setting external memory address areas which do not overlap each other for respective multiple caches; and providing virtual addresses with which the multiple processor cores access the multiple caches.

Abstract: An embodiment of the present disclosure may provide a multiply-accumulate operation method performed by a multiply-accumulate operation apparatus, the multiply-accumulate operation method including accumulating, by an accumulation register, a value within a preset bit value of a mantissa bitwidth in a result of an addition operation of a shifted first mantissa value and a shifted second mantissa value, determining, by an overflow counter, an overflow count based on an overflow value by which the result of the addition operation of the shifted first mantissa value and the shifted second mantissa value exceeds the preset bit value of the mantissa bitwidth, performing normalization and rounding based on the value accumulated in the accumulation register and the overflow count, and updating, by an exponent updater, the exponent using a normalized and rounded value.

Abstract: Disclosed herein is an Artificial Intelligence (AI) processor. The AI processor includes multiple NVM AI cores for respectively performing basic unit operations required for a deep-learning operation based on data stored in NVM; SRAM for storing at least some of the results of the basic unit operations; and an AI core for performing an accumulation operation on the results of the basic unit operation.

Abstract: A method for processing floating point operations in a multi-processor system including a plurality of single processor cores is provided. In this method, upon receiving a group setting for performing an operation, the plurality of single processor cores are grouped into at least one group according to the group setting, and a single processor core set as a master in the group loads an instruction for performing the operation from an external memory, and performs parallel operations by utilizing floating point units (FUPs) of all single processor cores in the group according to the instructions.

Abstract: Disclosed is a parallel processor. The parallel processor includes a processing element array including a plurality of processing elements arranged in rows and columns, a row memory group including row memories corresponding to rows of the processing elements, a column memory group including column memories corresponding to columns of the processing elements, and a controller to generate a first address and a second address, to send the first address to the row memory group, and to send the second address to the column memory group. The controller supports convolution operations having mutually different forms, by changing a scheme of generating the first address.

Abstract: Disclosed herein is an Artificial Intelligence (AI) processor. The AI processor includes multiple NVM AI cores for respectively performing basic unit operations required for a deep-learning operation based on data stored in NVM; SRAM for storing at least some of the results of the basic unit operations; and an AI core for performing an accumulation operation on the results of the basic unit operation.

Abstract: Disclosed herein is a heterogeneous system based on unified virtual memory. The heterogeneous system based on unified virtual memory may include a host for compiling a kernel program, which is source code of a user application, in a binary form and delivering the compiled kernel program to a heterogenous system architecture device, the heterogenous system architecture device for processing operation of the kernel program delivered from the host in parallel using two or more different types of processing elements, and unified virtual memory shared between the host and the heterogenous system architecture device.

Abstract: Disclosed herein is a method for outer-product-based matrix multiplication for a floating-point data type includes receiving first floating-point data and second floating-point data and performing matrix multiplication on the first floating-point data and the second floating-point data, and the result value of the matrix multiplication is calculated based on the suboperation result values of floating-point units.

Abstract: Disclosed is a transmitter which includes a channel driver that includes a pull-up transistor and a pull-down transistor connected between a power node and a ground node and outputs a voltage between the pull-up transistor and the pull-down transistor as a transmit signal, and a pre-driver that controls the pull-up transistor and the pull-down transistor in response to a driving signal and controls the channel driver such that the transmit signal is overshot at a rising edge of the driving signal and the transmit signal is undershot at a falling edge of the driving signal.

Abstract: A method and apparatus for allocating tasks through relocating kernel data based on a size of systolic array included in each of the plurality of processors, and relocating input feature map (IFM) data based on a number of the plurality of processors are provided.

Abstract: A method for ZQ calibration for a data transmission driving circuit of each memory die in a memory chip package in which memory dies are stacked, includes generating a reference current through a reference resistor connected between a power terminal supplying a power voltage of the data transmission driving circuit and a ground terminal and a first transistor that is diode-connected; supplying first currents corresponding to the reference currents to a pull-up driver of each memory die; performing ZQ calibration of a pull-up driver of a corresponding memory die by comparing a first voltage formed by each first current with a reference voltage formed by the reference current in each of the plurality of memory dies; and performing ZQ calibration of a pull-down driver of the corresponding memory die based on an output impedance of the ZQ calibrated pull-up driver in each of the memory dies.

Abstract: Disclosed herein is an apparatus for receiving a strobe signal. The apparatus may include an amplifier for amplifying a strobe signal input thereto, an offset generator for controlling the setting of a threshold for detecting a preamble signal by generating an offset for the amplifier, and a preamble detector for detecting a first preamble signal occurring at a point at which the amplified strobe signal is equal to or greater than the threshold and turning off the offset generator when the first preamble signal is detected.

Abstract: Provided is a memory interface device. A memory interface device, comprising: a DQS input buffer configured to receive input data strobe signals and output a first intermediate data strobe signal, the DQS input buffer providing a static offset; an offset control circuit configured to receive the first intermediate data strobe signal and output a second intermediate data strobe signal; and a duty adjustment buffer configured to receive the second intermediate data strobe signal and output a clean data strobe signal, wherein the offset control circuit provides a dynamic offset using the clean data strobe signal.

Abstract: Disclosed herein is an Artificial Intelligence (AI) processor. The AI processor includes multiple NVM AI cores for respectively performing basic unit operations required for a deep-learning operation based on data stored in NVM; SRAM for storing at least some of the results of the basic unit operations; and an AI core for performing an accumulation operation on the results of the basic unit operation.

Abstract: Disclosed herein is an AI accelerator. The AI accelerator includes processors, each performing a deep-learning operation using multiple threads; and a cache memory including an L0 instruction cache for providing instructions to the processors and an L1 cache mapped to the multiple areas of mapped memory.