Reconfigurable Computing¶
Computing by mapping functions onto a reusable hardware fabric whose logic or processing interconnections can be retargeted.
Core Idea¶
Reconfigurable computing maps computational work onto a hardware fabric whose implemented logic, processor-cell behavior or interconnections can be changed by loading another configuration. The configured hardware executes the work, but the same substrate can later be retargeted without fabricating a new fixed circuit. A general-purpose host is common, not mandatory; changing configuration during each execution is a narrower runtime strategy, not the defining requirement.[^ref-b042c1779125]
This differs from ordinary software configurability, which leaves the hardware computation path unchanged, and from a fixed ASIC, which cannot be retargeted in this fashion after fabrication. An FPGA is one possible fabric, not the whole architecture or the only fabric type.[ref-b042c1779125][ref-33ccc6742d46]
Scope of Application¶
The approach includes single-chip arrays and distributed FPGA fabrics. Putnam and colleagues' Catapult system mapped parts of Bing ranking into an interconnected FPGA pipeline while server software handled other stages; its measured throughput result applies to that deployment and comparison conditions.[^ref-059f4d8e6d90] Singh and colleagues' MorphoSys combined a reconfigurable processor-cell array, RISC controller and memory interface for multimedia tasks evaluated in simulation.[^ref-33ccc6742d46] These are unlike implementations of the same workload-to-retargetable-hardware relation.
Clarity¶
Separate three questions: what function is mapped, what hardware arrangement a configuration selects, and when another configuration is loaded. A device can remain configured for an entire run and still be reconfigurable; a claim of runtime reuse requires actual reloads during application execution. Host/fabric division and partial reconfiguration are design choices, not universal roles.[^ref-b042c1779125]
Manages Complexity¶
The abstraction organizes a design around workload mapping, fabric resources, configuration, hardware execution, data movement and resource policy. This prevents a fast hardware kernel from being mistaken for an end-to-end system gain: configuration and transfers may cost more than the mapped work saves. Catapult's server software and FPGA pipeline, for example, must be evaluated together rather than as an isolated chip.[ref-b042c1779125][ref-059f4d8e6d90]
Abstract Reasoning¶
Identify a workload portion suitable for hardware mapping, select fabric resources and a configuration, trace inputs and outputs across the system, and specify the reload policy. Compare total latency, throughput, energy or cost against a baseline under the same workload and measurement boundary. Runtime switching pays only when useful work between changes can amortize configuration and coordination overhead; that is an analytic test, not a universal measured gain.[^ref-b042c1779125]
Knowledge Transfer¶
The defining relation transfers from Catapult's datacenter FPGA fabric to MorphoSys's on-chip processor-cell array. Fabric granularity, interconnect, host coupling, reload schedule and performance result do not transfer unchanged. The workspace DAG proposal is unparented: live Reconfiguration describes theoretical reachability problems, while live Computer Architecture is a related broad field rather than a proven necessary typed genus. A portable reusable-substrate-specialization skeleton remains a future-prime question.
[^ref-b042c1779125]: Katherine Compton and Scott Hauck, “Reconfigurable Computing: A Survey of Systems and Software,” ACM Computing Surveys 34(2), 171–210 (2002), original scholarly survey. https://people.ece.uw.edu/hauck/publications/ConfigCompute.pdf . [^ref-059f4d8e6d90]: Andrew Putnam and colleagues, “A Reconfigurable Fabric for Accelerating Large-Scale Datacenter Services,” ISCA (2014), original full paper. Microsoft Research author-hosted PDF. [^ref-33ccc6742d46]: Hartej Singh and colleagues, “MorphoSys: A Reconfigurable Architecture for Multimedia Applications,” XI Brazilian Symposium on Integrated Circuit Design (1998), original full paper, DOI 10.1109/SBCCI.1998.715427. Original paper filed as a USPTO exhibit.
Neighborhood in Abstraction Space¶
Reconfigurable Computing sits in a moderately populated region (56th percentile for distinctiveness): it has near-neighbors but no dense thicket of look-alikes.
Family — Software & Systems Architecture (29 abstractions)
Nearest neighbors
- Position-Independent Code — 0.85
- Software-defined data center — 0.85
- Compute kernel — 0.85
- Reduced Instruction Set Computer — 0.85
- Instruction Set Architecture — 0.85
Computed from structural-signature embeddings · 2026-10-08