Abstract: A metallization arrangement for an integrated circuit is provided with a plurality of pads, such as bondpads and a dielectric layer. These pads are separated from each other by gaps. Reinforcing structures between the gaps mechanically reinforce the dielectric layer and reduce the potential for cracking, especially when a low k dielectric layer is employed.

Abstract: A memory implements a programmable physical address mapping that can change to reflect changing memory access patterns, observed or anticipated, to the memory. The memory employs address decode logic that can implement any of a variety of physical address mappings between physical addresses and corresponding memory locations. The physical address mappings may locate the data within one or more banks and rows of the memory so as to facilitate more efficient memory accesses for a given access pattern. The programmable physical address mapping employed by the hardware of the memory can include, but is not limited to, hardwired logic gates, programmable look-up tables or other mapping tables, reconfigurable logic, or combinations thereof. The physical address mapping may be programmed for the entire memory or on a per-memory region basis.

Abstract: A method and apparatus for state encoding of cache lines is described. Some embodiments of the method and apparatus support probing, in response to a first probe of a cache line in a first cache, a copy of the cache line in a second cache when the cache line is stale and the cache line is associated with a copy of the cache line stored in the second cache that can bypass notification of the first cache in response to modifying the copy of the cache line.

Abstract: An electronics chassis has many removable boards on sleds that are interconnected by a honeycomb interconnect structure. Interconnect boards in Y-planes and Z-planes are orthogonal to each other and form cells. Cooling air flows through the cells in an X direction, parallel to surfaces of the interconnect boards. The removable boards have connectors that mate with an edge of Z-divider interconnect boards. Fans blow air through the cells in the honeycomb structure unimpeded since no boards are perpendicular to the airflow. Notches in the rear of the Z-divider boards provide airflow equalization allowing closer spacing of fans to the honeycomb structure. A sled carrier honeycomb structure is placed in front of the honeycomb interconnect structure to guide sleds into position. Sled carrier dividers are offset from the Z-divider boards to allow removable boards to align with Z-divider boards in the Z-planes, parallel to airflow.

Abstract: The operating point of a processing unit is controlled based on the power consumption (i.e., the rate of energy consumption) associated with a workload, wherein low power consumption may indicate short-duration workloads with idle phases and high power consumption may indicate long, sustained workloads. Energy credits are accumulated while a drain rate of a battery is lower than a threshold drain rate and the energy credits are consumed while the drain rate is higher than the threshold drain rate. The operating point of the processing unit may be increased from a first operating point to a second operating point in response to the energy credits exceeding a first threshold. The operating point of the processing unit may be decreased from the second operating point to the first operating point in response to the energy credits falling below a second threshold.

Abstract: A method and apparatus using temperature margin to balance performance with power allocation. Nominal, middle and high power levels are determined for compute elements. A set of temperature thresholds are determined that drive the power allocation of the compute elements towards a balanced temperature profile. For a given workload, temperature differentials are determined for each of the compute elements relative the other compute elements, where the temperature differentials correspond to workload utilization of the compute element. If temperature overhead is available, and a compute element is below a temperature threshold, then particular compute elements are allocated power to match or drive toward the balanced temperature profile.

Abstract: Embodiments are described for a method and system of enabling updates from a clock controller to be sent directly to a predictive synchronizer to manage instant changes in frequency between transmit and receive clock domains, comprising receiving receive and transmit reference frequencies from a phase-locked loop circuit, receiving receive and transmit constant codes from a controller coupled to the phase-locked loop circuit, obtaining a time delay factor to accommodate phase detection between the transmit and receive clock domains, and calculating new detection interval and frequency information using the time delay factor, the reference frequencies, and the constant codes.

Abstract: A processing device is provided that includes a first, second and third precision operation circuit. The processing device further includes a shared, bit-shifting circuit that is communicatively coupled to the first, second and third precision operation circuits. A method is also provided for multiplying a first and second binary number including adding a first exponent value associated with the first binary number to a second exponent value associated with the second binary number and multiplying a first mantissa value associated with the first binary number to a second mantissa value associated with the second binary number. The method includes performing the exponent adding and mantissa multiplying substantially in parallel. The method further includes performing at least one of adding or subtracting a third binary number to the product. Also provided is a computer readable storage device encoded with data for adapting a manufacturing facility to create an apparatus.

Abstract: A processor includes a first shader engine and a second shader engine. The first shader engine is configured to process pixel shaders for a first subset of pixels to be displayed on a display device. The second shader engine is configured to process pixel shaders for a second subset of pixels to be displayed on the display device. Both the first and second shader engines are also configured to process general-compute shaders and non-pixel graphics shaders. The processor may also include a level-one (L1) data cache, coupled to and positioned between the first and second shader engines.

Abstract: Processing nodes in a 3D torus network topology are connected together via an interconnect that introduces at least one irregularity into the link connections between processing nodes of each ring of the network. Each processing node detects whether there is an irregularity in its links with adjacent processing nodes in a ring. As the sockets or other processing node interfaces of the interconnect are wired to introduce this irregularity and as the positions of the processing nodes within a given ring are relative to this irregularity, the physical location of the processing nodes can be determined based on correlations between physical locations of the sockets relative to the irregularity and the positions of the processing nodes relative to the irregularity. Thus, the relative position of a processing node in the ring can be used to identify the socket with which the processing node is coupled, thereby facilitating network management operations.

Abstract: The present invention provides a method and apparatus for detecting intrusions in a processor-based system. One embodiment of the method includes calculating a first checksum from first bits representative of instructions in a block of a program concurrently with executing the instructions. This embodiment of the method also includes issuing a security exception in response to determining that the first checksum differs from a second checksum calculated prior to execution of the block using second bits representative of instructions in the block when the second checksum is calculated.

Abstract: Various semiconductor chip solder bump and underbump metallization (UBM) structures and methods of making the same are disclosed. In one aspect, a method is provided that includes depositing a layer of a first metallic material on a semiconductor chip. The first layer has a first physical quantity. A layer of a second metallic material is deposited on the layer of the first metallic material. The second layer has a second physical quantity. The first and second layers are reflowed to form a solder structure with a desired ratio of the first metallic material to the second metallic material.

Abstract: A system and method for mapping an address space to a non-power-of-two number of memory channels. Addresses are translated and interleaved to the memory channels such that each memory channel has an equal amount of mapped address space. The address space is partitioned into two regions, and a first translation function is used for memory requests targeting the first region and a second translation function is used for memory requests targeting the second region. The first translation function is based on a first set of address bits and the second translation function is based on a second set of address bits.

Abstract: A memory and a method of storing data in a memory are provided. The memory comprises a memory block comprising data bits and additional bits. The memory includes logic which, when receiving a first command, writes data into the data bits of the memory block, wherein the data is masked according to a first input. The logic, in response to a second command, writes data into the data bits of the memory block and writes a second input into the additional bits of the memory block.

Abstract: Central processing units (CPUs) in computing systems manage graphics processing units (GPUs), network processors, security co-processors, and other data heavy devices as buffered peripherals using device drivers. Unfortunately, as a result of large and latency-sensitive data transfers between CPUs and these external devices, and memory partitioned into kernel-access and user-access spaces, these schemes to manage peripherals may introduce latency and memory use inefficiencies. Proposed are schemes to reduce latency and redundant memory copies using virtual to physical page remapping while maintaining user/kernel level access abstractions.

Abstract: A multilevel memory system includes a plurality of memories and a processor having a memory controller. The memory controller classifies each memory in accordance with a plurality of memory classes based on its level, its type, or both. The memory controller partitions a unified memory address space into contiguous address blocks and allocates the address blocks among the memory classes. In some implementations, the memory controller then can partition the address blocks assigned to each given memory class into address subblocks and interleave the address subblocks among the memories of the memory class.

Abstract: A shader unit is configured to provide an increased and dynamically changeable amount of ALU processing bandwidth. The shader unit includes a plurality of ALUs for processing pixel data according to a shader program. Each of the ALUs is configurable to be enabled and disabled. When disabled, the ALU is powered off, thereby reducing the power consumption of the shader unit. In one embodiment, the plurality of ALUs are logically configured into groups called ALU-pipes, each of which can be enabled and disabled. When an ALU-pipe is disabled, each ALU associated with the disabled ALU-pipe is disabled. The shader unit includes a sequencer that executes the shader program, determines the number of ALUs to be enabled, receives an input data stream of pixel data, assigns groups of pixel data to each enabled ALU, sends the assigned pixel data to their respective ALUs, and sends ALU instructions to the ALUs to process the received pixel data according to the shader program.

Abstract: Described are aggregation devices and methods for interconnecting server nodes. The aggregation device can include an input region, an output region, and a memory switch. The input region includes a plurality of input ports. The memory switch has a shared through silicon via (TSV) memory coupled to the input ports for temporarily storing data received at the input ports from a plurality of source devices. The output region includes a plurality of output ports coupled to the TSV memory. The output ports provide the data to a plurality of destination devices. A memory allocation system coordinates a transfer of the data from the source devices to the TSV memory. The output ports receive and process the data from the TSV memory independently of a communication from the input ports.

Abstract: Methods and systems are provided for graphics processing unit optimization via wavefront reforming including queuing one or more work-items of a wavefront into a plurality of queues of a compute unit. Each queue is associated with a particular processor within the compute unit. A plurality of work passes are performed. A determination is made which of the plurality of queues are below a threshold amount of work-items. Remaining one or more work-items from the queues with remaining ones of the work-items are redistributed to the below threshold queues. A subsequent work pass is performed. The, repeating of the determining, redistributing, and performing the subsequent work pass is done until all the queues are empty.

Abstract: A die-stacked memory device incorporates a data translation controller at one or more logic dies of the device to provide data translation services for data to be stored at, or retrieved from, the die-stacked memory device. The data translation operations implemented by the data translation controller can include compression/decompression operations, encryption/decryption operations, format translations, wear-leveling translations, data ordering operations, and the like. Due to the tight integration of the logic dies and the memory dies, the data translation controller can perform data translation operations with higher bandwidth and lower latency and power consumption compared to operations performed by devices external to the die-stacked memory device.