Abstract: Systems, apparatuses, and methods for addressing bank hotspotting are described. A computing system includes a memory controller with an arbiter for determining how to arbitrate access to one or more memory device(s) for received requests. The arbiter categorizes each request in a manner that helps to ensure fair virtual channel distribution across the banks of the memory device(s). The category system includes bank hotspotting functions to give banks that have more requests more chances to go over banks with fewer requests. The category system is implemented proportionally with more category credits given to banks with higher bank depths within the virtual channel.

Abstract: Aspects of the disclosure provide for an accelerator capable of accelerating data dependent, irregular, and/or memory-bound operations. An accelerator as described herein includes a programmable engine for efficiently executing computations on-chip that are dynamic, irregular, and/or memory-bound, in conjunction with a co-processor configured to accelerate operations that are predictable in computational load and behavior on the co-processor during design and fabrication.

Abstract: Systems, apparatuses, and methods for addressing bank hotspotting are described. A computing system includes a memory controller with an arbiter for determining how to arbitrate access to one or more memory device(s) for received requests. The arbiter categorizes each request in a manner that helps to ensure fair virtual channel distribution across the banks of the memory device(s). The category system includes bank hotspotting functions to give banks that have more requests more chances to go over banks with fewer requests. The category system is implemented proportionally with more category credits given to banks with higher bank depths within the virtual channel.

Abstract: An apparatus includes a memory circuit and a memory controller circuit. The memory controller circuit may include a write request queue. The memory controller circuit may be configured to receive a memory request to access the memory circuit and determine if the memory request includes a read request or a write request. A received read request may be scheduled for execution, while a received write request may be stored in the write request queue. The memory controller circuit may reorder scheduled memory requests based on achieving a specified memory access efficiency and based on a number of write requests stored in the write request queue.

Abstract: Techniques relating to arbitration in a memory controller are disclosed. In some embodiments, the memory controller is configured to transition between read turns and writes turn according to a turn schedule. In some embodiments, the memory controller also receives reports from circuitry requesting memory transactions and determines a current latency tolerance value based on the reports. In some embodiments, the memory controller is configured to switch from a write turn to a read turn prior to a scheduled switch based on the current latency tolerance meeting a threshold value.

Abstract: An apparatus includes a memory circuit and a memory controller circuit. The memory controller circuit may include a write request queue. The memory controller circuit may be configured to receive a memory request to access the memory circuit and determine if the memory request includes a read request or a write request. A received read request may be scheduled for execution, while a received write request may be stored in the write request queue. The memory controller circuit may reorder scheduled memory requests based on achieving a specified memory access efficiency and based on a number of write requests stored in the write request queue.

Abstract: A memory request management system may include a memory device and a memory controller. The memory controller may include a read queue, a write queue, an arbitration circuit, a read credit allocation circuit, and a write credit allocation circuit. The read queue and write queue may store corresponding requests from request streams. The arbitration circuit may send requests from the read queue and write queue to the memory device based on locations of addresses indicated by the requests. The read credit allocation circuit may send an indication of an available read credit to a request stream in response to a read request from the request stream being sent from the read queue to the memory device. The write credit allocation circuit may send an indication of an available write credit to a request stream in response to a write request from the request stream being stored at the write queue.

Abstract: An embodiment of an apparatus includes a memory circuit and a memory controller circuit. The memory controller circuit may include a write request queue. The memory controller circuit may be configured to receive a memory request to access the memory circuit and determine if the memory request includes a read request or a write request. A received read request may be scheduled for execution, while a received write request may be stored in the write request queue. The memory controller circuit may reorder scheduled memory requests based on achieving a specified memory access efficiency and based on a number of write requests stored in the write request queue.

Abstract: Techniques relating to arbitration in a memory controller are disclosed. In some embodiments, the memory controller is configured to transition between read turns and writes turn according to a turn schedule. In some embodiments, the memory controller also receives reports from circuitry requesting memory transactions and determines a current latency tolerance value based on the reports. In some embodiments, the memory controller is configured to switch from a write turn to a read turn prior to a scheduled switch based on the current latency tolerance meeting a threshold value.

Abstract: A memory request management system may include a memory device and a memory controller. The memory controller may include a read queue, a write queue, an arbitration circuit, a read credit allocation circuit, and a write credit allocation circuit. The read queue and write queue may store corresponding requests from request streams. The arbitration circuit may send requests from the read queue and write queue to the memory device based on locations of addresses indicated by the requests. The read credit allocation circuit may send an indication of an available read credit to a request stream in response to a read request from the request stream being sent from the read queue to the memory device. The write credit allocation circuit may send an indication of an available write credit to a request stream in response to a write request from the request stream being stored at the write queue.

Abstract: An embodiment of an apparatus includes a memory circuit and a memory controller circuit. The memory controller circuit may include a write request queue. The memory controller circuit may be configured to receive a memory request to access the memory circuit and determine if the memory request includes a read request or a write request. A received read request may be scheduled for execution, while a received write request may be stored in the write request queue. The memory controller circuit may reorder scheduled memory requests based on achieving a specified memory access efficiency and based on a number of write requests stored in the write request queue.

Abstract: Techniques relating to arbitration in a memory controller are disclosed. In some embodiments, the memory controller implements a per-bank priority-based arbitration scheme for different types of memory traffic (e.g., with different quality of service parameters). In some embodiments, the memory controller is configured to provide per-bank overrides to the arbitration scheme based on latency tolerance reported by one or more requesters sending a particular type of memory traffic. Various techniques disclosed herein may improve performance, improve fairness among different types of memory traffic, and/or reduce power consumption.

Abstract: Techniques relating to arbitration in a memory controller are disclosed. In some embodiments, the memory controller implements a per-bank priority-based arbitration scheme for different types of memory traffic (e.g., with different quality of service parameters). In some embodiments, the memory controller is configured to provide per-bank overrides to the arbitration scheme based on latency tolerance reported by one or more requesters sending a particular type of memory traffic. Various techniques disclosed herein may improve performance, improve fairness among different types of memory traffic, and/or reduce power consumption.

Abstract: A memory request management system may include a memory device and a memory controller. The memory controller may include a read queue, a write queue, an arbitration circuit, a read credit allocation circuit, and a write credit allocation circuit. The read queue and write queue may store corresponding requests from request streams. The arbitration circuit may send requests from the read queue and write queue to the memory device based on locations of addresses indicated by the requests. The read credit allocation circuit may send an indication of an available read credit to a request stream in response to a read request from the request stream being sent from the read queue to the memory device. The write credit allocation circuit may send an indication of an available write credit to a request stream in response to a write request from the request stream being stored at the write queue.

Abstract: A method and apparatus for aligning calibration segments for increased availability of a memory subsystem is disclosed. In one embodiment, a memory subsystem includes a memory and a memory controller coupled thereto via a number of independently operable channels (interfaces). The memory controller may convey on each of the channels at least one corresponding data strobe signal. The data strobe signal in each channel may be periodically calibrated. The memory controller may be configured to align the periodic calibrations in time so that they are performed concurrently instead of in a staggered manner. During the time the calibrations are performed on each channel, the memory may be unavailable for normal accesses.