Patent classifications
G06F2212/453
Apparatus and method for configurable redundant fuse banks
An apparatus is contemplated for storing and providing configuration data to an integrated circuit device, the apparatus has a fuse array and a plurality of cores. The fuse array is disposed on a die. The fuse array has a first plurality of semiconductor fuses and a second plurality of semiconductor fuses. The plurality of cores is disposed on the die, where each of the plurality of cores is coupled to the fuse array. The each of the plurality of cores includes array control, configured to access the first and second pluralities of fuses, and configured to process first states of the first plurality of semiconductor fuses and second states of the second plurality of semiconductor fuses according to contents of a configuration data register.
Apparatus and method for storage and decompression of configuration data
An apparatus includes a plurality of cores and a fuse array. The plurality of cores is disposed on a die. The fuse array is disposed on the die and is coupled to each of the plurality of cores, where the fuse array includes a plurality of semiconductor fuses that are programmed with compressed configuration data for the each of the plurality of cores, and where the each of the plurality of cores accesses and decompresses all of the compressed configuration data upon power-up/reset, for initialization of elements within the each of the plurality of cores.
Multi-core fuse decompression mechanism
An apparatus is contemplated for storing and decompressing configuration data in a multi-core microprocessor. The apparatus includes a shared fuse array and a plurality of microprocessor cores. The shared fuse array is disposed on a die and comprises a plurality of semiconductor fuses programmed with compressed configuration data. The plurality of microprocessor cores is also disposed on the die, where each of the plurality of microprocessor cores is coupled to the shared fuse array and is configured to access all of the compressed configuration data during power-up/reset, for initialization of elements within the each of the plurality of cores. The each of the plurality of cores have a reset controller that is configured to decompress the all of the compressed configuration data, and to distribute decompressed configuration data to initialize the elements.
Apparatus and method for extended cache correction
An apparatus includes a semiconductor fuse array, a cache memory, and a plurality of cores. The semiconductor fuse array is disposed on a die, into which is programmed the configuration data. The semiconductor fuse array has a first plurality of semiconductor fuses that is configured to store compressed cache correction data. The cache memory is disposed on the die. The plurality of cores is disposed on the die, where each of the plurality of cores is coupled to the semiconductor fuse array and the cache memory, and is configured to access the semiconductor fuse array upon power-up/reset, to decompress the compressed cache correction data, and to distribute decompressed cached correction data to initialize the cache memory.
Methods for increasing cache hit rates for neural networks
Systems, apparatuses, and methods for achieving higher cache hit rates for machine learning models are disclosed. When a processor executes a given layer of a machine learning model, the processor generates and stores activation data in a cache subsystem a forward or reverse manner. Typically, the entirety of the activation data does not fit in the cache subsystem. The processor records the order in which activation data is generated for the given layer. Next, when the processor initiates execution of a subsequent layer of the machine learning model, the processor processes the previous layer's activation data in a reverse order from how the activation data was generated. In this way, the processor alternates how the layers of the machine learning model process data by either starting from the front end or starting from the back end of the array.
Apparatus and method for compression of configuration data
An apparatus includes a device programmer, coupled to a plurality of semiconductor fuses disposed on a die, configured to program the plurality of semiconductor fuses with compressed configuration data for a plurality of cores disposed separately on the die. The device programmer has a virtual fuse array and a compressor. The virtual fuse array is configured to store the configuration data for the plurality of cores. The configuration data includes a plurality of data types. The compressor is coupled to the virtual fuse array and is configured to read the virtual fuse array, and is configured to compress the configuration data by employing a plurality of compression algorithms to generate the compressed configuration data, where the plurality of compression algorithms correspond to the plurality of data types.
Multi-indexed micro-operations cache for a processor
Various example embodiments for supporting a multi-indexed micro-operations cache (MI-UC) in a processor are presented. Various example embodiments for supporting an MI-UC in a processor may be configured to support an MI-UC in which, for a UC line of the MI-UC, multiple indexes into the UC line, for multiple sets of micro-operations (UOPs) stored in the UC line based on decoding of multiple instructions, are supported.