#nisq

The digital-analog quantum computing technology from Qilimanjaro promises an expedited practical advantage.

DAQC stands for Digital-Analog Quantum Computing.

Digital-analog quantum computing, or DAQC, is a hybrid approach in quantum technology that deliberately combines the benefits of digital and analogue quantum computation. Years before entirely digital roadmaps, DAQC seeks to develop more efficient, scalable quantum algorithms and offer a practical computing advantage over current noisy devices.

DAQC is achieved by fusing the precision of digital logic with the power and realism of analogue physics.

The mechanism of hybridisation

The fundamental DAQC model makes use of specific roles played by each computing paradigm:

Analogue Subsystems: These subsystems manage intricate multi-qubit interactions. Instead of using time-consuming gate sequences, analogue quantum computing continuously modifies the physical properties of the system to replicate actual quantum dynamics. This enables complex, many-body problems to be natively encoded into the device.

Digital Control: This is used for precise single-qubit local operations.

By combining digital control and analogue dynamics, DAQC conducts multi-qubit entangling operations as continuous analogue evolutions, replacing lengthy chains of discrete gates.

Getting Past NISQ Limitations

Near-Intermediate Scale Quantum (NISQ) technology failures are mostly caused by two-qubit gate errors, restricted coherence periods, and calibration overheads.

DAQC promptly addresses these challenges:

Reduced Error Rates: By replacing long chains of discrete gates with continuous analogue evolutions that carry out multi-qubit entangling operations, DAQC significantly lowers the overall error.

Faster Computation: Due to a reduced wall-clock period, calculations can be completed within the device's important coherence windows.

Cost Efficiency: By improving noise resilience and reducing circuit depth, calibration and runtime overheads are decreased. Because fewer repetitions are needed to achieve target accuracy, consumers ultimately pay less for cloud execution.

Before fully error-corrected quantum computers are widely used, Qilimanjaro says this hybrid architecture offers a feasible path to useful quantum computing.

Investigation and Use

The concept of DAQC has strong underlying support. In 2020, fundamental research developed universal DAQC techniques that demonstrated the use of a fixed, Ising-type analogue resource to interleave single-qubit rotations. Simulations showed that these DAQC circuits outperformed equally expressive all-digital circuits by a significant margin under comparable issue sizes and realistic noise settings.

Significantly, a digital–analog implementation of the Quantum Fourier Transform (QFT), the basis of Shor's prime factorisation approach, demonstrated better fidelity under actual noise in 2020 as opposed to a fully digital QFT. As the number of qubits increased, accuracy really increased as well. This finding implies improved scaling for techniques based on phase estimation.

In 2024, hardware-level comparisons on superconducting prototypes confirmed these findings. In terms of fidelities across representative single- and two-qubit noise channels, digital–analog realisations of QFT and phase estimation generally performed better than their digital-only equivalents. When researchers successfully paired a universal set of gates with a calibrated, chip-wide analogue evolution in superconducting devices in 2025, they further highlighted the scale and breadth that digital-analog computation offers, reaching beyond-classical regimes even with minimal analogue control.

Significant Progress in Quantum Machine Learning (QML) Techniques

DAQC is predicted to be advantageous for Quantum Machine Learning (QML) methods. This synergy arises because the digital layer allows for the rapid production of data-encoding states, while the initial analogue Hamiltonians can act as rich reservoirs or continuous-time feature maps.

This provides a substantial effective depth at a fixed gate count. Furthermore, by treating the evolution periods and qubit couplings as trainable parameters, the system may generate expressive machine learning models with fewer parameters and lower compilation overhead than fully digital QML approaches. In structured analogue dynamics, device noise may serve as implicit regularisation, improving trainability and reducing the issue of barren plateaus. Together, these features suggest improved learning capabilities and increased cost-efficiency for short-term quantum machine learning (QML) applications.

(Qilimanjaro's SpeQtrum) Implementation

Qilimanjaro's SpeQtrum integrated platform instantly incorporates the DAQC paradigm. This unified framework gives customers a single point of access through the usage of digital QPUs, CPUs, and GPUs as well as Qilimanjaro's differential analogue quantum architecture.

SpeQtrum allows users to easily switch between native analogue evolutions and gate-based operations while developing and executing digital-analog algorithms on the same superconducting quantum substrate. This unified architecture makes it possible to explore a wide range of application cases, such as machine learning, optimisation, and quantum simulation (especially for materials and chemistry), without the need for separate hardware or complex workflows.

Quantum Elements Reveals Constellation: The AI-Powered Platform that Closes the Gap Between Quantum Software and Hardware

A significant milestone that could transform quantum technology has been reached with the release of the ground-breaking Constellation platform from the LA-based quantum software business Quantum Elements. Constellation was designed to be an AI-Native platform. It promises to speed up the development of quantum software and hardware. In order to further the industry's development of practical, fault-tolerant quantum computing, the company seeks to eliminate some of the most persistent barriers.

The platform's purported market readiness is demonstrated by the fact that Quantum Elements has already secured significant backing and established significant partnerships. The company has received funding from QDNL Participations (Quantum Delta Netherlands) and the USC Viterbi School of Engineering. It has also established partnerships with important players in the market, such as Amazon and Rigetti.

The Revolution of AI: Streamlining Complicated Quantum Processes

The main innovation of Constellation is the deep integration of agentic artificial intelligence. This integration creates a development environment that is extremely powerful and genuinely intuitive by going beyond traditional code assistance. Constellation primarily assists users in developing, refining, and executing complex quantum applications through natural language cues.

Quantum Elements directly addresses the steep learning curve that is frequently linked to quantum computing. This field is renowned for necessitating an understanding of sophisticated physics concepts, specific quantum programming languages, and intricate hardware peculiarities. Constellation asserts that it offers a unique method of streamlining this procedure.

Through the use of agentic AI, the platform facilitates the rapid development of comprehensive simulation-driven machine learning models. The entry hurdle for developers, academics, and companies is significantly lowered by this capability. Instead of spending hours manually converting a conceptual approach into optimised, hardware-specific quantum code, a scientist may simply guide the platform in plain English. The AI agent then assumes responsibility for managing the complex process, constructing the necessary quantum circuit, establishing the simulation environment, and adjusting the crucial parameters.

Because it saves a lot of time and enables the AI to explore optimisation pathways that a human developer might otherwise overlook, this iterative, AI-native platform approach ultimately results in faster discovery and more efficient application development. Accelerating the iteration cycle from basic idea to simulation and, eventually, to an optimised prototype is a crucial aspect of the Constellation vision, positioning the platform to significantly outperform traditional quantum programming approaches.

Perhaps Constellation's most obvious feature is its own simulation back-end, especially for scholars researching quantum electronics. Quantum Elements claims that this back-end supports the biggest and most advanced noisy-qubit simulator to date. This advanced modelling capability is crucial since the industry is now defined by the Noisy Intermediate-Scale Quantum (NISQ) era, in which machines contain a finite number of qubits that are very prone to errors and noise. Eliminating this noise is currently the industry's largest problem.

Customers may authentically prototype quantum systems utilising the advanced simulator, which provides crucial utility, prior to investing in hardware access. This removes the expense and latency of physical machine time, enabling hardware manufacturers to thoroughly test new control sequences and designs and software developers to accurately assess how their applications will work on a range of architectures, including superconducting, ion-trap, and photonic systems.

The platform's diagnostic capability, which allows users to pause and analyse errors within the simulation environment, is its most significant feature. This function goes much beyond simple debugging by providing users with unmatched diagnostic information that allows them to pinpoint the exact moment and mechanism of error propagation within a quantum circuit. Users can systematically adapt their applications for certain hardware back-ends by identifying readout errors, gate integrity issues, or small noise channels. This process, dubbed noise-aware compilation, is necessary to improve gate quality and maximise the useful computing capability of existing NISQ devices.

The value of this functionality has been confirmed by well-known hardware manufacturers. The CEO of Rigetti Computing, Dr. Subodh Kulkarni, asserted that Constellation's AI-powered solutions had enhanced engineering. He highlighted the platform's significance in bridging the gap between theoretical models and actual performance by praising Quantum Elements' "deep insights in modelling superconducting qubit systems to improve gate fidelity."

The Road to Logical Qubits and Fault Tolerance

The primary goal of the Constellation platform's development is to expedite the advancement of better error correction techniques, which will ultimately result in the production of reliable logical qubits. Achieving logical qubits—sensitive quantum data encoded over several physical qubits to protect it from noise—is necessary to construct a truly global, large-scale quantum computer.

Designing, testing, and optimising robust error-correcting codes can be quite challenging; in some situations, hundreds or even thousands of physical qubits may be required to secure a single logical qubit. Constellation's ability to accurately simulate large-scale noisy quantum systems and perform thorough error analysis directly addresses this significant challenge. The method allows researchers to accurately model these vast, noisy systems, significantly accelerating the necessary trial-and-error process.

The AI component adds even more value by assisting in the discovery of optimal code layouts, enhanced decoding algorithms, and more efficient error-mitigation strategies tailored to the particulars of real hardware noise. This groundbreaking work is essential to the creation of fault-tolerant quantum devices that can address difficult problems like complex cryptography or molecular modelling.

Quantum Elements is uniquely positioned to lead this research, especially with its leadership team. The company was co-founded by Harvard Professor Amir Yacoby, Chief Scientific Officer Prof. Daniel Lidar, a renowned authority on quantum control and error correction from the University of Southern California, and CEO Dr. Izhar Medalsy, Ph.D.

The company's technology is focused not only on developing quantum programs but also on developing the infrastructure and software layer that have a deep understanding of the hardware's physics in order to move the industry away from the current NISQ phase and towards full quantum usefulness.

To sum up

The unveiling of Constellation marks a sea change in the advancement of artificial intelligence and quantum computing. Quantum Elements is tackling the hardware-software mismatch head-on with an AI-native platform interface, an advanced noisy-qubit simulator, and precise error analysis tools. According to Kris Kaczmarek, Investment Director at QDNL Participations, the platform is anticipated to emerge as a premier instrument for developing quantum applications.

In 2027, PsiQuantum expects photons to power a million-qubit supercomputer.

The Alpha System

California startup PsiQuantum wants to develop a million-qubit fault-tolerant quantum computer by 2027. Quantum computing could transform chemistry and materials research if this aim is reached. Four British university researchers launched the company in 2016 to build a silicon photonics-based optical quantum computer. It has raised $1.7 billion. The “Alpha System,” PsiQuantum's first fully working prototype, is being assembled at its new Milpitas, California plant. By building on networking and photonics technology, the company may reach critical scale faster than competitors using “matter-based” solutions like superconducting qubits, trapped ions, or neutral atoms.

Scaling: Fault-Tolerance Required

From the start, PsiQuantum focused on building a full-scale, fault-tolerant quantum computer. While early industrial expectations focused on tiny, “noisy intermediate-scale quantum” (NISQ) computers performing significant work without error correction, comprehensive fault tolerance is now accepted for utility.

According to cofounder and chief scientific officer Pete Shadbolt, this fundamental premise drove the search for optical approaches. Cooling, control, communication, and manufacturability must be addressed to reach millions of qubits for error correction. Photonics may solve these difficulties better than competitors.

A New Cryogenics/Connectivity Method

Matter-based qubits are sensitive to radiation and temperature fluctuations because they must be chilled to almost absolute zero. Photons can act as qubits at ambient temperature since they resist radiation and heat.

Cryogenic temperatures are still needed because PsiQuantum's technology uses superconducting photon detectors between 2 and 4 kelvins. It's easier to achieve these temperatures. PsiQuantum's server-rack-sized cryogenic cabinets can hold 250 chips, unlike superconducting qubits' dilution freezers, which hold one or two chips. Linde's cryoplant will cool three of these huge cabinets in Milpitas.

The heat and radiation resistance of photons allows control electronics to be placed near qubits, simplifying system design. Communication is simplified by qubits' ability to be transported over telecom fiber. Recently, the company sent qubits over 250 meters of fiber with 99.7% fidelity.

Making Mass Production with Semiconductors

Manufacturing is one of the biggest obstacles to large-scale quantum computing because most systems are unique. For this, PsiQuantum created a commercial chip production method using silicon-photonics technology.

Shadbolt notes that the company was founded to manufacture millions of high-maturity devices using the trillions of dollars invested in the semiconductor industry over the past 50 years. Despite its superconducting photon detectors and ultrafast optical switches, Global Foundries is commercially producing thousands of PsiQuantum devices in Malta, New York.

“Flighty Photons” Technical Challenges

Despite these advantages, photonics has significant downsides. The linear optics technology makes photon generation nondeterministic, therefore quick fault tolerance is needed, according to Simon Devitt of the Center for Quantum Software and Information at the University of Technology, Sydney.

Gate operations in PsiQuantum fail 25–50% of the time. The business utilizes “multiplexing” to try many photon-generation methods and pick the best ones, but this only partially solves the problem. The remaining gate failures need error correction. Devitt believes that gate failures have taken up a large portion of the error-correction budget, leaving little room for other faults.

Laser loss is the second leading cause of errors after gate failures. This loss depends on waveguides, photon detectors, and optical switches. Despite data showing ready waveguides and detectors, the company's switches have high losses.

Shadbolt remains convinced that purity and manufacturing are the real issues, not material science in general. He believes thousands of little, gradual improvements in chip geometry, design, and manufacturing accuracy will lead to success.

Not Seeking Supremacy, Testing System

Building the Alpha System in Milpitas is the first major test of the business's design. Vice president of system architecture Mercedes Gimeno-Segovia said these exploratory testing will not involve quantum techniques. Unlike other companies who used NISQ prototypes to establish quantum supremacy on “toy problems,” PsiQuantum believes NISQ machines behave too differently from fault-tolerant ones to provide useful information.

The Alpha System is used to assess if the system's behavior matches the company's theoretical models, which is crucial for designing large-scale systems.

Quantum convolutional neural networks improve picture categorisation and feature processing.

Scientists have developed a novel quantum convolutional neural network (QCNN) technology that could improve picture categorisation efficiency and accuracy. Quantum physics is used to solve challenging pattern recognition issues that challenge conventional computer methods.

This study is led by Shaswata Mahernob Sarkar, Sheikh Iftekhar Ahmed, and colleagues from the University of Rochester and Bangladesh University of Engineering and Technology. Their research uses a parallel-mode QCNN and selective feature re-encoding to improve feature processing and classification accuracy.

Image Processing with Quantum Technology

Quantum CNN for image classification

Noisy Intermediate-Scale Quantum (NISQ) devices are advancing quantum machine learning. It has many uses, including image recognition. Despite their success, standard convolutional neural networks (CNNs) are computationally expensive and resource-intensive, especially for big and complicated datasets. This inherent limitation of standard CNNs has spurred research into Quantum Convolutional Neural Networks (QCNNs), which have superior representational capacity and computing efficiency. QCNNs use quantum notions like entanglement and superposition to improve image categorisation.

New Feature Extraction Methods

The researchers introduce two major breakthroughs in this groundbreaking study:

Selective Feature Re-Encoding Strategy: This novel method forces quantum circuits to prioritise an input image's most informative features. By selecting and encoding only the most important data, this method increase signal-to-noise ratio and reduce processing load. The quantum system uses this targeted strategy to explore Hilbert space, the complicated vector space that describes all quantum system states, to discover the optimal feature processing solutions.

A New Parallel QCNN Architecture: This complicated approach integrates PCA and autoencoder features into a single training scheme.

PCA, a popular dimensionality reduction method, identifies the most important variances in data.

In contrast, autoencoders are neural networks that train to efficiently extract important properties from compressed input data. By combining PCA and autoencoders in a quantum framework, the research team hopes to increase classification performance by creating a more precise and trustworthy feature representation.

Thorough Verification and Excellent Results

These methods were thoroughly tested using MNIST and Fashion-MNIST, two popular photo datasets. The number of qubits, the quantum equivalent of classical bits, and the quantum circuit depth affect the QCNN's performance. Investigations were crucial to showing this. The importance of optimisation and fine-tuning in QCNN design is highlighted.

The jointly optimised parallel QCNN architecture consistently outperformed both individual QCNN models and conventional ensemble approaches. Merging PCA and Autoencoders within the quantum framework gives the QCNN a more complete and trustworthy feature representation of input images, which improves performance.

Common quantum computing topics like orthogonal matrix decomposition and two-qubit gate circuit topologies underpin this study. Using strong quantum simulation and machine learning software tools and frameworks like TensorFlow Quantum and PennyLane helped with development and testing. This study's code and data are freely available in the quantum machine learning community to encourage repeatability and collaboration.

Introducing Future Quantum Developments

Over time, the study team hopes to increase the adoption of their QCNN design. They will test its applicability in more difficult picture classification tasks like object recognition and image segmentation. They may also consider employing more advanced quantum algorithms to boost performance. The development of hardware-efficient Quantum Convolutional Neural Networks (QCNNs) architectures for near-term quantum devices will also be prioritised, moving this technology towards real-world deployment.

New Tomography Methods for Quantum State Description Cut Measurements and Improve Fidelity

QST

QST reconstructs an unknown quantum state from experimental observations, enabling quantum technologies to reach their full potential and ensure quantum instrument efficiency. Many quantum tasks require an accurate mathematical representation of the quantum state.

Conventional quantum state tomography is hindered by statistical noise, experimental errors, and the exponential increase in experimental demands and resource requirements as qubits increase. Full quantum state tomography is often needed to access nonlinear properties, even if other features may not require it.

The Breakthrough: Less Measurement, More Fidelity

Researchers improved quantum state tomography for three-qubit states by using a unique method that only requires 17 measurements, compared to the usual 63 Pauli measurements. In the challenging context of quantum computers, H. Talath, B. P. Govindaraja, and colleagues reported their achievement, which will ease the description of complex quantum systems.

The primary breakthrough is based on a theoretical prediction: a careful set of partial observations can explain most three-qubit pure states. Pure quantum systems have known states. This method contrasts with standard tomography, which requires many more observations to create the density matrix, a mathematical description of the quantum state.

Open-source quantum computing platform ibm_osaka achieved this reduced protocol with its 127-qubit processor. During experiments, they replicated a three-qubit W state and its two-qubit marginals. Its two-qubit subsystems consistently surpassed standard, full three-qubit tomography in W state fidelity, a measure of how closely the recreated state resembles the original. This confirms the theoretical benefit of recreating states from subsystems.

NISQ Device Benefits from Noisiness

Due to Noisy Intermediate-Scale Quantum (NISQ) device restrictions, this subsystem-based technique is practicable. NISQ devices, which have a finite number of qubits, are error-prone. Measurement numbers must be reduced to reduce error and improve result reliability. By focused on relevant subsystems, the unique approach allows for more frequent and deep analysis and more dependable and precise results, especially for noisy quantum systems.

Dealing with Noise with SIC-POVMs

Another quantum tomography study characterises three-qubit states with amplitude-damping noise and other environmental perturbations. This study emphasises the importance of correctly describing pure three-qubit states in these situations for information processing and quantum computing. Even with noise, generalised single-qubit Symmetric Informationally Complete Positive Operator-Valued Measures (SIC-POVMs) measurement probabilities can effectively identify the quantum state.

This section analyses the impact of amplitude-damping noise on the quantum state and measurement technique. Any qubit entangled with a noise-affected qubit is also affected by the noise. This shows how multi-qubit systems' qubits are connected and how it affects noisy state reconstruction.

In a noise-free scenario, only 10 of 64 possible outputs from SIC-POVM measurements can fully reconstruct a three-qubit pure quantum state and determine its unknown coefficients. This theory defines the relationship between measurement probabilities and noise characteristics to improve quantum state reconstruction in the presence of environmental disruptions.

Benefits and Challenges of SIC-POVMs

Since all tomographic data is in one experimental run, SIC-POVMs are conceptually advantageous. Quantum state tomography of neural networks has this advantage. Thus, when system size expands with Pauli tomography, it is empirically far more cost-effective to repeat the same measurement setup numerous times than to swap configurations exponentially.

Real-world applications are affected by this trait. Since each run includes complete tomographic information, the experimenter can end the method at any time, unlike other quantum state tomography-related systems that require at least one run per measurement setup to collect adequate information. Although SIC-POVMs have several benefits, they require consecutive measurements and precise settings, making them challenging to use.

Wider Effects

Quantum-Hybrid SVMs Quantum Kernels Revolutionise Critical Infrastructure Anomaly Detection

New study presented today shows that Quantum-Hybrid Support Vector Machines (QSVMs) may detect anomalies in Industrial Control Systems (ICS), advancing critical infrastructure cybersecurity. The research paper “Quantum-Hybrid Support Vector Machines for Anomaly Detection in Industrial Control Systems” by Cultice, Hassan Onim, Giani, and Thapliyal found that QSVMs outperformed classical kernel techniques with a 13.3% higher F1 score and 91.023% better kernel alignment.

To fight against modern cyberattacks, critical infrastructure needs anomaly detection. They manage physical operations that generate enormous amounts of data, making input fraud harder to detect. Modern security concerns go beyond SCADA warnings, prompting the search for  machine learning  models.

Quantum Machine Learning (QML) uses quantum kernels' expressive feature spaces to address this growing cybersecurity challenge. Quantum-Hybrid Support Vector Machines (QSVMs) use projected quantum kernel functions to transfer data into a higher-dimensional space for traditional SVM analysis. Similarities and differences in data may be too computationally expensive for standard computers to discover. Quantum computing is used for kernel fidelity computations, whereas data pre-processing and SVM components are normally handled to reduce noise and resource utilisation. Key study findings include: Accuracy: Quantum-Hybrid Support Vector Machines (QSVMs) improve F1 scores by 13.3% across all datasets, outperforming classic kernel methods. The F1 score, which measures a test's memory and precision, indicates a stronger capacity to identify odd behaviour. QSVMs outperformed standard methods in kernel-target alignment by 91.023%. In a multi-dimensional "feature space," this enhanced alignment separates normal and anomalous data more efficiently, reducing false positives and negatives and improving anomaly detection system reliability. A higher kernel target alignment suggests a “quantum advantage”. Noise Resilience: QSVM kernels had a 0.98% error rate in IBMQ hardware simulations. Despite this inaccuracy lowering classification metrics by 1.57%, QSVMs outperformed classical methods, proving their durability and usefulness. The study examined QSVM performance using datasets from real cyber-physical systems as hydropower producing (HAI), water distribution (WADI), and water treatment (SWaT). The Belis et al. kernel and the basic U2-gate “2DoF” kernel performed best, showing that less sophisticated quantum models can achieve efficiency by preventing overfitting. Quantum-Hybrid Support Vector Machines (QSVMs) show potential, however the paper emphasises their disadvantages and urges further research. These include testing QSVM integration with other machine learning approaches, transfer learning, and scaling QSVM models to handle larger and more complex datasets. The study shows that quantum hardware still has long queues and limited computing time for large datasets in the NISQ (Noisy Intermediate-Scale Quantum) era. Future research includes deeper studies employing improved quantum computing resources, more dependable and effective quantum algorithms, and moving beyond simulations to genuine quantum hardware. This study shows that Quantum-Hybrid Support Vector Machines (QSVMs) can give ICS an edge in anomaly detection, improving critical infrastructure cybersecurity.

To conclude