Asynchronous Pipeline Merging via Bitwise Vector Arbitration
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Chiplet systems face timing issues due to varying latencies in parallel processing pipelines, leading to inefficient pipeline throughput and increased power consumption, particularly in memory operations where some requests are satisfied by cache while others require access to off-die memory.
Innovation Solution
The implementation of vectors to represent transactions in parallel paths, allowing for asynchronous and independent completion logging, followed by a bitwise AND operation to identify completed transactions, which are then managed by a wide, fast, fair Round-Robin arbiter to ensure fair handling and progression through the pipeline.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If parallel processing pipelines are used to handle memory requests, then processing capacity is improved, but timing synchronization becomes difficult due to varying latencies
Solution Approach 1:
The patent segments the memory request processing into multiple independent parallel pipelines, each capable of handling requests asynchronously. By dividing the monolithic processing into separate pipelines with independent timing, the system achieves high processing capacity while each pipeline maintains its own timing synchronization, resolving the contradiction between parallel processing capability and timing coordination.
2Device complexity
If synchronous pipeline merging is used, then timing control is simplified, but pipeline throughput decreases due to stalls waiting for slowest path
Solution Approach 1:
The patent implements dynamic pipeline merging where the merging mechanism adapts to the completion status of individual pipelines. Instead of forcing synchronous completion, the system dynamically selects which pipeline results to merge based on availability, allowing fast pipelines to contribute results without waiting for slow pipelines. This dynamic approach maintains simple timing control while dramatically improving throughput by eliminating stall cycles.
3Stability of the object's composition
If pipelines wait for slowest path completion, then timing alignment is maintained, but power consumption increases due to idle resource wastage
Solution Approach 1:
The patent employs periodic arbitration and selective merging where pipelines are evaluated at regular intervals for completion status. Instead of continuous waiting, the system periodically checks which pipelines have completed and merges those results immediately. This periodic action maintains timing alignment for merged outputs while allowing individual pipelines to proceed independently, reducing idle power consumption significantly.
Data Source
AI summary
Devices and techniques for asynchronous pipeline merging are described herein. An apparatus, includes a memory controller, which includes merge circuitry; where the memory controller chiplet is configured to perform operations including those to: perform a bitwise logical operation on a first logging bit vector and a second logging bit vector to obtain a result vector, wherein the first logging bit vector is associated with a first pipeline and the second logging bit vector is associated with a second pipeline, and wherein bits in respective index positions of the first and second logging bit vectors represent transactions; select a completed transaction from the result vector using a round-robin technique; and forward the completed transaction from the set of completed transactions to an output pipeline.