QCORE: A Quantum-Computing-Oriented,Resource-Efficient Runtime Architecture with Lifecycle Coordination

arXiv:2608.06875 · quant-ph, cs.AR · Submitted 2026-08-07 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Quantum Radio. Generated commentary on the latest quantum physics and condensed matter papers.

Kai: I'm Kai.

Mira: And I'm Mira, here with Lev.

Kai: We're looking at the paper: "QCORE: A Quantum-Computing-Oriented,Resource-Efficient Runtime Architecture with Lifecycle Coordination".

Mira: Here's the gist of the paper. Scalable quantum processors require control, readout, feedback, calibration, and error correction to coexist under bounded latency and shared-resource constraints <ref:2608.06875#pg2> QCORE (Quantum-ControlOriented RealTime Execution) is a QPU-side digital control reference architecture positioned between the Host and a platform-specific analog/mixed-signal front end <ref:2608.06875#pg4> This article presents QCORE (Quantum-ControlOriented RealTime Execution), a QPU-side digital control reference architecture positioned between the Host and a platform-specific analog/mixed-signal front end <ref:2608.06875#pg2> The principal contributions of this work are as follows <ref:2608.06875#pg5> A QPU-side digital control reference architecture that connects to the Host software stack through a unified runtime package and isolates task management, shared resources, realtime execution, and closed-loop services into distinct hardware partitions <ref:2608.06875#pg5> A dual-output readout interface comprising a fast-result sideband and a Measurement Packet <ref:2608.06875#pg5> The former supports hard-real-time same-round feedback, whereas the latter unifies traceable measurement and service data with timestamps, resource identifiers, and configuration versions <ref:2608.06875#pg5> A versioned safe-state mechanism combined with tile-local closure <ref:2608.06875#pg5> The Tile-local QEC kernel handles deadline-critical actions in the current round, while long-term parameters are written to a shadow configuration and atomically committed <ref:2608.06875#pg5> Architectural Partitioning QCORE partitions the complete system into the Host, QPUside digital control, a platform-specific analog/mixed-signal front end, and the QPU <ref:2608.06875#pg4> The Host performs high-level compilation, mapping, global optimization, and model training <ref:2608.06875#pg4> QCORE separates task management, shared resources, hard-real-time execution, and long-timescale services into four hardware partitions <ref:2608.06875#pg2> The Control Center admits and dispatches the runtime package received from the Host <ref:2608.06875#pg4> The Real-Time Execution Engine reads the RTQS program, templates, and active configuration, and uses the RTQS, Timing Engine, and Gate-to-Pulse Mapper (GPM) to generate control operations while performing readout and local feedback <ref:2608.06875#pg4> Data Flow Mechanisms The readout path in QCORE produces two parallel outputs <ref:2608.06875#pg4> The fast-result sideband from the Fast Classifier is delivered directly to the Fast Feedback Unit for same-round conditional branching, active reset, and frame update <ref:2608.06875#pg4> The Measurement Packet is routed to services, the result buffer, and the Host <ref:2608.06875#pg4> A service output can enter the feedback path only after deadline, freshness, and version checks <ref:2608.06875#pg4> A complete Measurement Packet is not a prerequisite for same-round feedback; a result that misses the action window can only be logged or applied to a later round <ref:2608.06875#pg4> Service Organization The Closed-Loop Service Subsystem instantiates the Calibration Service and ErrorCorrection Service on a common service skeleton <ref:2608.06875#pg5> The Calibration Service performs sequence/sweep generation, feature reduction, parameter estimation, and candidate generation for Rabi, Ramsey, readout, crosstalk, and drift workloads <ref:2608.06875#pg5> The Error-Correction Service organizes syndromeextraction profiles, long-term histories, decoder models, and policy evaluation but is not placed on the current-round deadline path <ref:2608.06875#pg5> The Tile-local QEC kernel performs detector generation, lightweight decoding, deadline checking, and Pauli-frame/correction actions <ref:2608.06875#pg5> System Evaluation Metrics The evaluation compares architectural mechanisms and does not replace cycle-accurate RTL, PPA analysis, or hardware measurement <ref:2608.06875#pg4> At a load of 0.8, the mean latencies of FIFO, WRR, and real-time priority are 3.11Lmax, 2.30Lmax, and 1.40Lmax respectively <ref:2608.06875#pg4> Closed-loop operation reduces the mean frequency error by 83.2%±0.8% and lowers the state-assignment error at maximum readout drift from 10.39%±0.54% to 5.37%±0.29% <ref:2608.06875#pg2> No unsafe acceptance or mixed-version observation is observed in 100,000 configuration transactions <ref:2608.06875#pg2> Scalability and Concurrency QCORE scales local control and readout through replicable Control/Readout Tiles <ref:2608.06875#pg5> The Qubit Map, Topology/Coupler, Channel Binding, Readout Group, and Service Map jointly constrain the physical resources and subscriptions of each tile <ref:2608.06875#pg5> Local conflicts are checked within a tile; cross-tile resources are coordinated through global reservation or distributed tokens <ref:2608.06875#pg5> The sustainable tile-count ratio is 2.08, based on trace-derived provisioning envelope rather than a silicon saturation throughput <ref:2608.06875#pg4> QCORE does not merely collect control, calibration, and error-correction blocks <ref:2608.06875#pg5> Its dualoutput readout interface separates the shortest feedback path from the traceable service path <ref:2608.06875#pg2> Together, these mechanisms establish explicit boundaries among task, measurement, event, and configuration-update flows <ref:2608.06875#pg5> The functional and system-level behavioral models show that runtime packages can drive Rabi–Ramsey and RB workloads through admission, execution, packetization, and aggregation <ref:2608.06875#pg4> Safe-Commit and tile-local closure constrain configuration consistency and global-service pressure, respectively <ref:2608.06875#pg5> Future work will refine the transaction/event model into cycle-accurate RTL, complete synthesis and physical implementation, and validate sustained multiservice operation with a physical analog front end and QPU <ref:2608.06875#pg4> REFERENCES

1: Alexander T, Kanazawa N, Egger D J, et al.

Kai: Qiskit pulse: programming quantum computers through the cloud with pulses

J: .

Mira: Quantum Science & Technology, 2020, 5(4): 044006 <ref:2608.06875#pg2>

2: Cross A, Javadi-Abhari A, Alexander T, et al.

Kai: OpenQASM 3: A broader and deeper quantum assembly language

J: .

Paper summary: Kai: So we talked about this QCORE paper where they propose this digital control reference architecture that sits between your host software and the platform specific analog front end to handle all those coexisting requirements.

Mira: The core thesis is that they partition the system into four hardware partitions to separate task management, shared resources, hard real time execution, and long timescale services.

Lev: From my side I see this architecture as a way to manage the overhead of running closed loop services on top of real hardware without it all collapsing under latency or resource contention.

Kai: And they introduce this dual output readout interface which gives you a fast result sideband for immediate feedback and a measurement packet for traceable service data with versioning and timestamps.

Mira: That dual path is what allows them to separate the short feedback path from the longer traceable service path, which is important because they have to be handled differently.

Lev: It's smart design because you need one pipeline for immediate reaction and another pipeline for logging things that might take a while to process.

Kai: And then they detail how they use a versioned safe state mechanism with tile local closure so the current round actions are handled quickly while long term parameters are updated in a shadow configuration.

Mira: That way they avoid background services corrupting the real time state consistency because of those version and safe point conditions checks.

Lev: It’s about making sure that the most urgent things get done correctly right now, and then managing the updates to other parts of the system later without causing a mess.

Kai: So basically, they are giving us a blueprint for how to structure these complex quantum control systems so that everything coexists under bounded latency and shared resource constraints.

Conclusion: Kai: Thinking about this QCORE architecture, the title really points to how they are making runtime execution and resource efficiency a central focus for scalable quantum processors.

Mira: I think it suggests that the way we structure these systems needs to move beyond just collecting blocks and instead focus on how those blocks interact under real constraints.

Lev: It implies that if we want to build bigger machines, we can't just stack components; we need a coordination layer like this digital reference architecture to manage the lifecycle of everything.

Kai: So what this means for us is that future work needs to focus on refining these transaction event models into cycle accurate hardware and validating sustained multiservice operation with a physical analog front end and the actual QPU.

Mira: The authors flag that they need to complete synthesis and physical implementation for their full vision, which means we still have a lot of hardware validation left before this is fully realized.

Lev: That’s fair because moving from behavioral models to actual silicon means dealing with all those physical realities like power consumption and thermal noise.

Kai: It boils down to building systems where the control, calibration, and error correction aren't just separate things but are explicitly coordinated under strict timing rules.

Heyue Li, Yanshu Guo, Qichun Liu, Tiefu Li, Zhihua Wang, Hanjun Jiang

School of Integrated Circuits, Tsinghua University · School of Integrated Circuits, Shanghai Jiao Tong University · Beijing Academy of Quantum Information Sciences

quant-ph, cs.AR

Submitted: 2026-08-07

Updated: 2026-11-09

Comments: 18 pages, 10 figures

License: http://creativecommons.org/licenses/by-nc-sa/4.0/

Importance score: 96/100

The gist: Scalable quantum processors require control, readout, feedback, calibration, and error correction to coexist under bounded latency and shared-resource constraints <ref:2608.06875#pg2> QCORE…

Key concepts

QCORE
Quantum-ControlOriented RealTime Execution is a QPU-side digital control reference architecture. It sits between the host software and the physical analog front end of a quantum processor, managing task execution, shared resources, real-time operations, and closed-loop services in distinct hardware partitions.
Dual-Output Readout Interface
This interface provides two parallel data streams from qubit measurements. The first is a fast-result sideband used for immediate same-round feedback actions like conditional branching. The second is a Measurement Packet, which carries traceable service data, including timestamps and configuration versions, for later logging or service processing.
Tile-local QEC Kernel
This kernel handles time-critical error correction tasks within the current round. It performs detector generation and lightweight decoding while checking deadlines. Long-term parameters are stored in a shadow configuration and only committed atomically, ensuring deadline-critical actions are prioritized.

Terminology

Summary

Scalable quantum processors require control, readout, feedback, calibration, and error correction to coexist under bounded latency and shared-resource constraints <ref:2608. p. 0

Where the paper says it

cn, jianghanjun@tsinghua. ABSTRACT Scalable quantum processors require control, readout, feedback, calibration, and error correction to coexist under bounded latency and shared-resource constraints, whereas existing platforms typically optimize only a subset of these capabilities.

06875#pg2> QCORE (Quantum-ControlOriented RealTime Execution) is a QPU-side digital control reference architecture positioned between the Host and a platform-specific analog/mixed-signal front end <ref:2608. p. 0
Where the paper says it

ABSTRACT Scalable quantum processors require control, readout, feedback, calibration, and error correction to coexist under bounded latency and shared-resource constraints, whereas existing platforms typically optimize only a subset of these capabilities. This article presents QCORE (Quantum-ControlOriented Real-Time Execution), a QPU-side digital control reference architecture positioned between the Host and a platform-specific analog/mixed-signal front end.

06875#pg4> This article presents QCORE (Quantum-ControlOriented RealTime Execution), a QPU-side digital control reference architecture positioned between the Host and a platform-specific analog/mixed-signal front end <ref:2608. p. 0
Where the paper says it

ABSTRACT Scalable quantum processors require control, readout, feedback, calibration, and error correction to coexist under bounded latency and shared-resource constraints, whereas existing platforms typically optimize only a subset of these capabilities. This article presents QCORE (Quantum-ControlOriented Real-Time Execution), a QPU-side digital control reference architecture positioned between the Host and a platform-specific analog/mixed-signal front end.

06875#pg2> The principal contributions of this work are as follows <ref:2608. p. 1
Where the paper says it

QCORE complements upper-layer software frameworks and specialized front ends by focusing on hardware execution and service organization after a task reaches the QPU side. Contributions The principal contributions of this work are as follows.

06875#pg5> A QPU-side digital control reference architecture that connects to the Host software stack through a unified runtime package and isolates task management, shared resources, realtime execution, and closed-loop services into distinct hardware partitions <ref:2608. p. 1
Where the paper says it

Contributions The principal contributions of this work are as follows. A QPU-side digital control reference architecture that connects to the Host software stack through a unified runtime package and isolates task management, shared resources, realtime execution, and closed-loop services into distinct hardware partitions.

06875#pg5> A dual-output readout interface comprising a fast-result sideband and a Measurement Packet <ref:2608. p. 1
Where the paper says it

A QPU-side digital control reference architecture that connects to the Host software stack through a unified runtime package and isolates task management, shared resources, realtime execution, and closed-loop services into distinct hardware partitions. A dual-output readout interface comprising a fast-result sideband and a Measurement Packet.

06875#pg5> The former supports hard-real-time same-round feedback, whereas the latter unifies traceable measurement and service data with timestamps, resource identifiers, and configuration versions <ref:2608. p. 1
Where the paper says it

A dual-output readout interface comprising a fast-result sideband and a Measurement Packet. The former supports hard-real-time same-round feedback, whereas the latter unifies traceable measurement and service data with timestamps, resource identifiers, and configuration versions.

06875#pg5> A versioned safe-state mechanism combined with tile-local closure <ref:2608. p. 1
Where the paper says it

The Calibration Service and Error-Correction Service consequently reuse a common service-control skeleton while retaining service-specific kernels. A versioned safe-state mechanism combined with tile-local closure.

06875#pg5> The Tile-local QEC kernel handles deadline-critical actions in the current round, while long-term parameters are written to a shadow configuration and atomically committed <ref:2608. p. 1
Where the paper says it

A versioned safe-state mechanism combined with tile-local closure. The Tile-local QEC kernel handles deadline-critical actions in the current round, while long-term parameters are written to a shadow configuration and atomically committed

06875#pg5>

Architectural Partitioning

QCORE partitions the complete system into the Host, QPUside digital control, a platform-specific analog/mixed-signal front end, and the QPU <ref:2608.06875#pg4> The Host performs high-level compilation, mapping, global optimization, and model training <ref:2608. p. 2

Where the paper says it

System Boundary QCORE partitions the complete system into the Host, QPUside digital control, a platform-specific analog/mixed-signal front end, and the QPU. The Host performs high-level compilation, mapping, global optimization, and model training.

06875#pg4> QCORE separates task management, shared resources, hard-real-time execution, and long-timescale services into four hardware partitions <ref:2608. p. 0
Where the paper says it

This article presents QCORE (Quantum-ControlOriented Real-Time Execution), a QPU-side digital control reference architecture positioned between the Host and a platform-specific analog/mixed-signal front end. QCORE separates task management, shared resources, hard-real-time execution, and long-timescale services into four hardware partitions.

06875#pg2> The Control Center admits and dispatches the runtime package received from the Host <ref:2608. p. 2
Where the paper says it

2, QCORE comprises a Control Center, Shared Support Infrastructure, Real-Time Execution Engine, and Closed-Loop Service Subsystem for task management, controlled sharing, hard-real-time execution, and longtimescale services, respectively. The Control Center admits and dispatches the runtime package received from the Host.

06875#pg4> The Real-Time Execution Engine reads the RTQS program, templates, and active configuration, and uses the RTQS, Timing Engine, and Gate-to-Pulse Mapper (GPM) to generate control operations while performing readout and local feedback <ref:2608. p. 2
Where the paper says it

The Control Center admits and dispatches the runtime package received from the Host. The Real-Time Execution Engine reads the RTQS program, templates, and active configuration, and uses the RTQS, Timing Engine, and Gate-to-Pulse Mapper (GPM) to generate control operations while performing readout and local feedback.

06875#pg4>

Data Flow Mechanisms

The readout path in QCORE produces two parallel outputs <ref:2608.06875#pg4> The fast-result sideband from the Fast Classifier is delivered directly to the Fast Feedback Unit for same-round conditional branching, active reset, and frame update <ref:2608. p. 2

Where the paper says it

2 produces two parallel outputs. The fast-result sideband from the Fast Classifier is delivered directly to the Fast Feedback Unit for same-round conditional branching, active reset, and frame update.

06875#pg4> The Measurement Packet is routed to services, the result buffer, and the Host <ref:2608. p. 2
Where the paper says it

The fast-result sideband from the Fast Classifier is delivered directly to the Fast Feedback Unit for same-round conditional branching, active reset, and frame update. The Measurement Packet is routed to services, the result buffer, and the Host.

06875#pg4> A service output can enter the feedback path only after deadline, freshness, and version checks <ref:2608. p. 2
Where the paper says it

The Measurement Packet is routed to services, the result buffer, and the Host. A service output can enter the feedback path only after deadline, freshness, and version checks, while a long-term parameter can take effect only through a safe shadow-to-active configuration commit.

06875#pg4> A complete Measurement Packet is not a prerequisite for same-round feedback; a result that misses the action window can only be logged or applied to a later round <ref:2608. p. 5
Where the paper says it

The Fast Feedback Unit applies deadline, freshness, and version checks to a sideband result or eligible service event before generating an RTQS branch, GPM rewrite, active reset, conditional pulse, or frame update. A complete Measurement Packet is not a prerequisite for same-round feedback; a result that misses the action window can only be logged or applied to a later round.

06875#pg4>

Service Organization

The Closed-Loop Service Subsystem instantiates the Calibration Service and ErrorCorrection Service on a common service skeleton <ref:2608.06875#pg5> The Calibration Service performs sequence/sweep generation, feature reduction, parameter estimation, and candidate generation for Rabi, Ramsey, readout, crosstalk, and drift workloads <ref:2608. p. 5

Where the paper says it

Outputs are routed by type to the event network, Frame Table, shadow configuration, or Host. The Calibration Service performs sequence/sweep generation, feature reduction, parameter estimation, and candidate generation for Rabi, Ramsey, readout, crosstalk, and drift workloads.

06875#pg5> The Error-Correction Service organizes syndromeextraction profiles, long-term histories, decoder models, and policy evaluation but is not placed on the current-round deadline path <ref:2608. p. 5
Where the paper says it

Computationally intensive fitting may remain at the Host. The Error-Correction Service organizes syndromeextraction profiles, long-term histories, decoder models, and policy evaluation but is not placed on the current-round deadline path.

06875#pg5> The Tile-local QEC kernel performs detector generation, lightweight decoding, deadline checking, and Pauli-frame/correction actions <ref:2608. p. 5
Where the paper says it

The Error-Correction Service organizes syndromeextraction profiles, long-term histories, decoder models, and policy evaluation but is not placed on the current-round deadline path. The Tile-local QEC kernel performs detector generation, lightweight decoding, deadline checking, and Pauli-frame/correction actions.

06875#pg5>

System Evaluation Metrics

The evaluation compares architectural mechanisms and does not replace cycle-accurate RTL, PPA analysis, or hardware measurement <ref:2608.06875#pg4> At a load of 0. p. 0

Where the paper says it

Transaction-level, event-driven, and quantumbehavioral models are used for evaluation. At a background load of 0.

8, the mean latencies of FIFO, WRR, and real-time priority are 3. p. 8
Where the paper says it

At a load of 0. 8, the mean latencies of FIFO, WRR, and real-time priority are 3.

11Lmax, 2.30Lmax, and 1.40Lmax respectively <ref:2608. p. 2
Where the paper says it

QCORE System Architecture As shown in Fig. 2, QCORE comprises a Control Center, Shared Support Infrastructure, Real-Time Execution Engine, and Closed-Loop Service Subsystem for task management, controlled sharing, hard-real-time execution, and longtimescale services, respectively.

06875#pg4> Closed-loop operation reduces the mean frequency error by 83. p. 0
Where the paper says it

8, the P99 latency of the shared Measurement Packet/Event feedback path is (1. Closed-loop operation reduces the mean frequency error by 83.

2%±0.8% and lowers the state-assignment error at maximum readout drift from 10. p. 0
Where the paper says it

Closed-loop operation reduces the mean frequency error by 83. 8% and lowers the state-assignment error at maximum readout drift from 10.

39%±0.54% to 5.37%±0.29% <ref:2608.06875#pg2> No unsafe acceptance or mixed-version observation is observed in 100,000 configuration transactions <ref:2608. p. 0
Where the paper says it

8% and lowers the state-assignment error at maximum readout drift from 10. No unsafe acceptance or mixed-version observation is observed in 100,000 configuration transactions, and Tile-local QEC reduces modeled global-boundary demand and yields a 2.

06875#pg2>

Scalability and Concurrency

QCORE scales local control and readout through replicable Control/Readout Tiles <ref:2608.06875#pg5> The Qubit Map, Topology/Coupler, Channel Binding, Readout Group, and Service Map jointly constrain the physical resources and subscriptions of each tile <ref:2608. p. 5

Where the paper says it

Local gates, readout, active reset, and round-critical QEC actions close within the tile, while cross-tile gates, barriers, and frame updates enter the hierarchical interconnect. The Qubit Map, Topology/Coupler, Channel Binding, Readout Group, and Service Map jointly constrain the physical resources and subscriptions of each tile.

06875#pg5> Local conflicts are checked within a tile; cross-tile resources are coordinated through global reservation or distributed tokens <ref:2608. p. 5
Where the paper says it

The system can scale from one tile with centralized SRAM to multiple tiles with banked SRAM and hierarchical NoC/Event/Time-Sync fabrics. Local conflicts are checked within a tile; cross-tile resources are coordinated through global reservation or distributed tokens.

06875#pg5> The sustainable tile-count ratio is 2. p. 0
Where the paper says it

Quantum-Computing System Requirements Scaling a quantum processor requires its control, readout, calibration, and feedback infrastructure to scale concurrently. Conventional room-temperature instrument chains facilitate experimental development but encounter cable-count, synchronization, data-movement, and feedback-latency limitations under multichannel operation and long unattended runs.

08, based on trace-derived provisioning envelope rather than a silicon saturation throughput <ref:2608.06875#pg4>

Conclusion

QCORE does not merely collect control, calibration, and error-correction blocks <ref:2608.06875#pg5> Its dualoutput readout interface separates the shortest feedback path from the traceable service path <ref:2608. p. 0

Where the paper says it

QCORE separates task management, shared resources, hard-real-time execution, and long-timescale services into four hardware partitions. A fast-result sideband closes same-round feedback, a Measurement Packet provides a traceable measurement and service interface, and a common service-control skeleton, Tile-local QEC, and versioned safe-point commit organize calibration, error correction, and long-term state updates.

06875#pg2> Together, these mechanisms establish explicit boundaries among task, measurement, event, and configuration-update flows <ref:2608. p. 2
Where the paper says it

This boundary isolates pulse-level hard-real-time paths from software management and long-timescale services while permitting controlled sharing of storage, configuration, and event resources. In this article, deterministic execution denotes a behavioral architectural constraint: an admitted operation carries an explicit timestamp or relative timing constraint, resource conflicts are checked before issue, and the real-time stream cannot be arbitrarily interrupted by firmware, background services, or ordinary NoC transactions.

06875#pg5> The functional and system-level behavioral models show that runtime packages can drive Rabi–Ramsey and RB workloads through admission, execution, packetization, and aggregation <ref:2608. p. 5
Where the paper says it

Rabi–Ramsey calibration and randomized benchmarking (RB) are selected to evaluate multistage state update and batched circuit execution. The transaction model covers package/ABI validation, task admission, RTQS/GPM expansion, Measurement Packet routing, and result aggregation.

06875#pg4> Safe-Commit and tile-local closure constrain configuration consistency and global-service pressure, respectively <ref:2608. p. 1
Where the paper says it

A versioned safe-state mechanism combined with tile-local closure. The Tile-local QEC kernel handles deadline-critical actions in the current round, while long-term parameters are written to a shadow configuration and atomically committed

06875#pg5> Future work will refine the transaction/event model into cycle-accurate RTL, complete synthesis and physical implementation, and validate sustained multiservice operation with a physical analog front end and QPU <ref:2608. p. 0
Where the paper says it

The control electronics must therefore provide deterministic timing, low-latency feedback, and isolation from background services. This work retains high-level compilation, databases, model training, and global scheduling at the Host; moves template expansion, readout processing, conditional feedback, and closed-loop state maintenance into a digital layer close to the QPU; and delegates physical-signal conversion to a platformspecific analog/mixed-signal front end.

06875#pg4>

REFERENCES

[1] Alexander T, Kanazawa N, Egger D J, et al. Qiskit pulse: programming quantum computers through the cloud with pulses[J]. p. 1

Where the paper says it

Related Work Qiskit Pulse, OpenQASM 3, and pulse-level optimization provide low-level control and dynamic-circuit representations [1]-[3], while CUDA-Q supports heterogeneous CPU/GPU/QPU programming [16], [17]. Quantum-centric supercomputing and tightly coupled HPC-QPU architectures further emphasize coordination between QPUs and HPC resources [18], [19].

Quantum Science & Technology, 2020, 5(4): 044006 <ref:2608. p. 0
Where the paper says it

1 QCORE: A Quantum-Control-Oriented RealTime Execution Architecture with Extensible Closed-Loop Services and Shared AI Acceleration Heyue Li1 , Yanshu Guo2 , Qichun Liu3 , Tiefu Li1,3 , Zhihua Wang1 , Hanjun Jiang1 1School of Integrated Circuits, Tsinghua University, Beijing, China 2School of Integrated Circuits, Shanghai Jiao Tong University, Shanghai, China 3Beijing Academy of Quantum Information Sciences, Beijing, China Corresponding author: Yanshu Guo and Hanjun Jiang (email: guoyanshu@sjtu. cn, jianghanjun@tsinghua.

06875#pg2>

[2] Cross A, Javadi-Abhari A, Alexander T, et al. OpenQASM 3: A broader and deeper quantum assembly language[J]. p. 1

Where the paper says it

Related Work Qiskit Pulse, OpenQASM 3, and pulse-level optimization provide low-level control and dynamic-circuit representations [1]-[3], while CUDA-Q supports heterogeneous CPU/GPU/QPU programming [16], [17]. Quantum-centric supercomputing and tightly coupled HPC-QPU architectures further emphasize coordination between QPUs and HPC resources [18], [19].

ACM Transactions on Quantum Computing, 2022, 3(3): 1-50 <ref:2608. p. 0
Where the paper says it

INTRODUCTION A. Quantum-Computing System Requirements Scaling a quantum processor requires its control, readout, calibration, and feedback infrastructure to scale concurrently.

06875#pg2>

[3] Niu S, Todri-Sanial A. Effects of dynamical decoupling and pulselevel optimizations on ibm quantum computers[J]. p. 0

Where the paper says it

1 QCORE: A Quantum-Control-Oriented RealTime Execution Architecture with Extensible Closed-Loop Services and Shared AI Acceleration Heyue Li1 , Yanshu Guo2 , Qichun Liu3 , Tiefu Li1,3 , Zhihua Wang1 , Hanjun Jiang1 1School of Integrated Circuits, Tsinghua University, Beijing, China 2School of Integrated Circuits, Shanghai Jiao Tong University, Shanghai, China 3Beijing Academy of Quantum Information Sciences, Beijing, China Corresponding author: Yanshu Guo and Hanjun Jiang (email: guoyanshu@sjtu. cn, jianghanjun@tsinghua.

IEEE Transactions on Quantum Engineering, 2022, 3: 1-10 <ref:2608. p. 0
Where the paper says it

8% and lowers the state-assignment error at maximum readout drift from 10. No unsafe acceptance or mixed-version observation is observed in 100,000 configuration transactions, and Tile-local QEC reduces modeled global-boundary demand and yields a 2.

06875#pg2>

[4] Stefanazzi L, Treptow K, Wilcer N, et al. The QICK (Quantum Instrumentation Control Kit): Readout and control for qubits and detectors[J]. p. 1

Where the paper says it

These efforts define the software and system environment above QCORE but do not address realtime execution, resource isolation, or state commit in QPUside digital hardware. At the control-hardware level, QICK integrates waveform generation and readout, Fu et al.

Review of Scientific Instruments, 2022, 93(4) <ref:2608.06875#pg2>

[5] Fu X, Rol M A, Bultink C C, et al. An experimental microarchitecture for a superconducting quantum processor[C]//Proceedings of the 50th Annual IEEE/ACM International Symposium on Microarchitecture. p. 0

Where the paper says it

Quantum-Computing System Requirements Scaling a quantum processor requires its control, readout, calibration, and feedback infrastructure to scale concurrently. Conventional room-temperature instrument chains facilitate experimental development but encounter cable-count, synchronization, data-movement, and feedback-latency limitations under multichannel operation and long unattended runs.

2017: 813-825 <ref:2608.06875#pg2>

[6] Xue X, Patra B, van Dijk J P G, et al. CMOS-based cryogenic control of silicon quantum circuits[J]. p. 0

Where the paper says it

1 QCORE: A Quantum-Control-Oriented RealTime Execution Architecture with Extensible Closed-Loop Services and Shared AI Acceleration Heyue Li1 , Yanshu Guo2 , Qichun Liu3 , Tiefu Li1,3 , Zhihua Wang1 , Hanjun Jiang1 1School of Integrated Circuits, Tsinghua University, Beijing, China 2School of Integrated Circuits, Shanghai Jiao Tong University, Shanghai, China 3Beijing Academy of Quantum Information Sciences, Beijing, China Corresponding author: Yanshu Guo and Hanjun Jiang (email: guoyanshu@sjtu. cn, jianghanjun@tsinghua.

Nature, 2021, 593(7858): 205-210 <ref:2608.06875#pg2>

[7] Guo Y, Li Y, Huang W, et al. A polar-modulation-based cryogenic qubit state controller in 28nm bulk CMOS[C]//2023 IEEE International Solid-State Circuits Conference (ISSCC). IEEE, 2023: 508-510 <ref:2608.06875#pg2>

[8] Lennon D T, Moon H, Camenzind L C, et al. Efficiently measuring a quantum device using machine learning[J]. p. 2

Where the paper says it

This separation prevents background services from corrupting real-time state consistency. The functional flow, real-time isolation, closed-loop update, and scaling mechanisms are evaluated using transaction/event models, a three-level transmon master-equation model, and a dispersive-IQ readout model.

npj Quantum Information, 2019, 5(1): 79 <ref:2608. p. 0
Where the paper says it

1 QCORE: A Quantum-Control-Oriented RealTime Execution Architecture with Extensible Closed-Loop Services and Shared AI Acceleration Heyue Li1 , Yanshu Guo2 , Qichun Liu3 , Tiefu Li1,3 , Zhihua Wang1 , Hanjun Jiang1 1School of Integrated Circuits, Tsinghua University, Beijing, China 2School of Integrated Circuits, Shanghai Jiao Tong University, Shanghai, China 3Beijing Academy of Quantum Information Sciences, Beijing, China Corresponding author: Yanshu Guo and Hanjun Jiang (email: guoyanshu@sjtu. cn, jianghanjun@tsinghua.

06875#pg2>

[9] Darulová J, Pauka S J, Wiebe N, et al. Autonomous tuning and charge-state detection of gate-defined quantum dots[J]. p. 1

Where the paper says it

demonstrated cryogenic CMOS controllers [4]-[7]. Autonomous tuning, real-time feedback, and QEC experiments have further established the importance of online calibration, conditional actions, and repeated syndrome processing [8]-[15].

Physical Review Applied, 2020, 13(5): 054005 <ref:2608. p. 5
Where the paper says it

The Fast Feedback Unit applies deadline, freshness, and version checks to a sideband result or eligible service event before generating an RTQS branch, GPM rewrite, active reset, conditional pulse, or frame update. A complete Measurement Packet is not a prerequisite for same-round feedback; a result that misses the action window can only be logged or applied to a later round.

06875#pg2>

[10] Ristè D, Bultink C C, Lehnert K W, et al. Feedback control of a solid-state qubit using high-fidelity projective measurement[J]. p. 0

Where the paper says it

The control electronics must therefore provide deterministic timing, low-latency feedback, and isolation from background services. This work retains high-level compilation, databases, model training, and global scheduling at the Host; moves template expansion, readout processing, conditional feedback, and closed-loop state maintenance into a digital layer close to the QPU; and delegates physical-signal conversion to a platformspecific analog/mixed-signal front end.

Physical review letters, 2012, 109(24): 240502 <ref:2608. p. 0
Where the paper says it

The control electronics must therefore provide deterministic timing, low-latency feedback, and isolation from background services. This work retains high-level compilation, databases, model training, and global scheduling at the Host; moves template expansion, readout processing, conditional feedback, and closed-loop state maintenance into a digital layer close to the QPU; and delegates physical-signal conversion to a platformspecific analog/mixed-signal front end.

06875#pg2>

[11] Córcoles A D, Takita M, Inoue K, et al. Exploiting dynamic quantum circuits in a quantum algorithm with superconducting qubits[J]. p. 0

Where the paper says it

1 QCORE: A Quantum-Control-Oriented RealTime Execution Architecture with Extensible Closed-Loop Services and Shared AI Acceleration Heyue Li1 , Yanshu Guo2 , Qichun Liu3 , Tiefu Li1,3 , Zhihua Wang1 , Hanjun Jiang1 1School of Integrated Circuits, Tsinghua University, Beijing, China 2School of Integrated Circuits, Shanghai Jiao Tong University, Shanghai, China 3Beijing Academy of Quantum Information Sciences, Beijing, China Corresponding author: Yanshu Guo and Hanjun Jiang (email: guoyanshu@sjtu. cn, jianghanjun@tsinghua.

Physical Review Letters, 2021, 127(10): 100501 <ref:2608. p. 0
Where the paper says it

The control electronics must therefore provide deterministic timing, low-latency feedback, and isolation from background services. This work retains high-level compilation, databases, model training, and global scheduling at the Host; moves template expansion, readout processing, conditional feedback, and closed-loop state maintenance into a digital layer close to the QPU; and delegates physical-signal conversion to a platformspecific analog/mixed-signal front end.

06875#pg4>

[12] Kelly J, Barends R, Fowler A G, et al. State preservation by repetitive error detection in a superconducting quantum circuit[J]. p. 1

Where the paper says it

Related Work Qiskit Pulse, OpenQASM 3, and pulse-level optimization provide low-level control and dynamic-circuit representations [1]-[3], while CUDA-Q supports heterogeneous CPU/GPU/QPU programming [16], [17]. Quantum-centric supercomputing and tightly coupled HPC-QPU architectures further emphasize coordination between QPUs and HPC resources [18], [19].

Nature, 2015, 519(7541): 66-69 <ref:2608.06875#pg4>

[13] Suppressing quantum errors by scaling a surface code logical qubit[J]. Nature, 2023, 614(7949): 676-681 <ref:2608.06875#pg4>

[14] Liu J, Lee Y, Xu Y, et al. A Scalable Open-Source QEC System with Sub-Microsecond Decoding-Feedback Latency[J]. p. 8

Where the paper says it

System-Level Quantitative Evaluation An event-driven simulator maps Calibration, QEC, and RB traces onto the real-time execution, shared interconnect, closed-loop service, configuration-commit, and tile-local feedback paths and couples these paths to the transmon/readout model. Latency is normalized to 𝐿𝑚𝑎𝑥, the service time of a maximum-length transaction.

arXiv preprint arXiv:2603.16203, 2026 <ref:2608.06875#pg2>

[15] Yang X, Sun X, Wu Z, et al. Real-time Surface-Code Error Correction Using an FPGA-based Neural-Network Decoder[J]. p. 2

Where the paper says it

The Real-Time Execution Engine reads the RTQS program, templates, and active configuration, and uses the RTQS, Timing Engine, and Gate-to-Pulse Mapper (GPM) to generate control operations while performing readout and local feedback. The Shared Support Infrastructure provides the NoC/SRAM subsystem, event network, configuration versioning, Hardware-Abstraction Tables, and an optional neural processing unit (NPU).

arXiv preprint arXiv:2605.04892, 2026 <ref:2608.06875#pg2>

[16] Rallis K, Liliopoulos I, Varsamis G D, et al. Interfacing quantum computing systems with high-performance computing systems: An overview[J]. p. 0

Where the paper says it

INTRODUCTION A. Quantum-Computing System Requirements Scaling a quantum processor requires its control, readout, calibration, and feedback infrastructure to scale concurrently.

arXiv preprint arXiv:2509.06205, 2025 <ref:2608.06875#pg2>

[17] Ganguly S. Hybrid Classical--Quantum Learning for Space Based Data Centers: A CUDA-Q Study of Variational and Photonic Backends[J]. p. 1

Where the paper says it

Related Work Qiskit Pulse, OpenQASM 3, and pulse-level optimization provide low-level control and dynamic-circuit representations [1]-[3], while CUDA-Q supports heterogeneous CPU/GPU/QPU programming [16], [17]. Quantum-centric supercomputing and tightly coupled HPC-QPU architectures further emphasize coordination between QPUs and HPC resources [18], [19].

2026 <ref:2608.06875#pg2>

[18] Seelam S, Chow J M, Córcoles A, et al. Reference architecture of a quantum-centric supercomputer[J]. p. 0

Where the paper says it

ABSTRACT Scalable quantum processors require control, readout, feedback, calibration, and error correction to coexist under bounded latency and shared-resource constraints, whereas existing platforms typically optimize only a subset of these capabilities. This article presents QCORE (Quantum-ControlOriented Real-Time Execution), a QPU-side digital control reference architecture positioned between the Host and a platform-specific analog/mixed-signal front end.

arXiv preprint arXiv:2603.10970, 2026 <ref:2608.06875#pg2>

[19] Caldwell S A, Khazraee M, Agostini E, et al. Platform architecture for tight coupling of high-performance computing with quantum processors[J]. p. 0

Where the paper says it

ABSTRACT Scalable quantum processors require control, readout, feedback, calibration, and error correction to coexist under bounded latency and shared-resource constraints, whereas existing platforms typically optimize only a subset of these capabilities. This article presents QCORE (Quantum-ControlOriented Real-Time Execution), a QPU-side digital control reference architecture positioned between the Host and a platform-specific analog/mixed-signal front end.

arXiv preprint arXiv:2510.25213, 2025 <ref:2608.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Close-loop operation reduces the mean frequency error by 83.2% plus 0.8 percent and lowers state-assignment error at maximum readout drift from 10.39 percent plus 0.54 percent to 5.37 percent plus 0.29 percent, demonstrating improved stability through closed-loop feedback on the QPU side.

  2. The dual-output readout interface provides a fast result sideband for hard real-time same-round feedback and a Measurement Packet that unifies traceable measurement and service data with timestamps, resource identifiers, and configuration versions.

  3. The Tile-local QEC kernel handles deadline critical actions in the current round while long term parameters are written to a shadow configuration and atomically committed only after version dependency checks, preventing background services from corrupting real time state consistency.

  4. The system can scale by replicating Control/Readout Tiles where local gates, readout, active reset, and round critical QEC actions close within the tile whenever possible.

  5. A platform independent runtime package is translated by the Host backend into QCORE objects, allowing upper layer frameworks like CUDA-Q or Qiskit to utilize the architecture without being tied to specific physical front ends.

  6. Traffic class isolation in the NoC QoS mechanism separates real time and background queues by traffic class and arbitrates at transaction boundaries, limiting interference from DMA and service traffic.

  7. The Calibration Service performs sequence sweep generation, feature reduction, parameter estimation, and candidate generation for various workloads like Rabi or Ramsey sweeps to mitigate frequency drift.

  8. SafeCommit ensures no unsafe acceptance or mixed version observation occurs in 100,000 configuration transactions by requiring validation of expected version and safepoint conditions before switching active configurations.

Abstract

As quantum processors scale, online calibration, quantum error correction (QEC), and measurement-driven feedback require continuous classical processing and parameter updates. Sharing on-chip resources reduces hardware cost, but computation completion, the last parameter read, and result consumption occur at different times. Their dependencies must be managed separately while preserving quantum operations' real-time requirements. We present QCORE, a quantum-computing-oriented, resource-efficient runtime architecture that coordinates these lifecycles for safe resource reuse and parameter consistency. QCORE models resource occupancy, parameter protection, and consumer dependencies as independent lifecycles and provides three mechanisms: pre-use checks and parameter binding to verify timing, identity, and dependencies; access-aware protection through value snapshots, version references, or active-configuration windows; and joint admission of resource reservations and parameter installation. After request submission, hardware autonomously handles admission, protection, installation, and reclamation, reducing fine-grained Host coordination and providing a foundation for one Host to manage multiple controllers. We implement synthesizable RTL and validate FPGA timing at 100 MHz. Independent release reduces whole-program execution time by 16.83% in Ramsey calibration and QEC replay. Joint admission increases on-time completions from 5/15 to 15/15 under contention across five arrival phases. Under identical external deadlines, a shared NN backend reduces LUT usage by 12.28%-13.81% and DSP usage by 39.39% compared with two statically assigned backends. These results show that lifecycle coordination exploits inter-task timing slack for on-chip sharing, increasing service capacity under limited resources.

Sources

Related papers