Bilevel QAOA yields fully connected QBM at single layer
Inner loop minimizes energy while outer loop tunes Hamiltonian parameters, hitting 0.9559 target probability noiseless and staying dominant,
cs.ET
Emerging Technologies
Covers approaches to information processing (computing, communication, sensing) and bio-chemical analysis based on alternatives to silicon CMOS-based technologies, such as nanoscale electronic, photonic, spin-based, superconducting, mechanical, bio-chemical and quantum technologies (this list is not exclusive). Topics of interest include (1) building blocks for emerging technologies, their scalability and adoption in larger systems, including integration with traditional technologies, (2) modeling, design and optimization of novel devices and systems, (3) models of computation, algorithm design and programming for emerging technologies.
sort pith recommended most recent
This paper reviews recent advancements in mm-wave oscillators below 100 GHz and…
A literature review summarizing design approaches, performance metrics, challenges, and emerging techniques for mm-wave and sub-THz/THz…
· “Recent Advances in mm-Wave and Sub-THz/THz Oscillators for FutureG Technologies”
Reported 415.8x compiler gap shrinks to 3.26x under unified test
A shared format and evaluator reveal how much of the reported gap was assumption, not compiler quality.
· “Practical Insights into Fair Comparison and Evaluation Frame for Neutral-Atom Compilers”
PoLaRIS claims a single flow from device design to DRV-free layouts
Merges fabrication-aware inverse design with automated placement and routing to cut manual PIC layout work.
Muscle signals lift exercise-form scoring past vision-only AI
New benchmark pairs video, pose, and muscle signals so AI can name the exact mistake, not just issue a score.
· “MyoMechanix: Biomechanically-Grounded Compositional Skilled Activity Understanding and Coaching”
One stretchable wire carries data and reports its own stretch
Reading amplitude and digital pattern lets one liquid-metal line carry data and sense strain without extra sensors.
· “Simultaneous Digital Communication and Deformation Sensing over a Single Stretchable Interconnect”
ReRAM resists 100k stray pulses
Non-linearity below 0.005 keeps stray pulses from shifting weights; simulated training reaches 89-95%.
· “Update Disturbance-Resilient Analog ReRAM Crossbar Arrays for In-Memory Deep Learning Accelerators”
Compact model ties ReRAM symmetry point to equilibrium defect state
Simulations show pulsed symmetry point matches quasi-static equilibrium defect state; an I-V sweep can predict it.
One ACC setting change lifts highway throughput 58.6%
Switching from cruise to follow mode at the safe time gap also cuts speed oscillations by 39.7 percent.
· “A unified dynamical modeling framework for cruise control and adaptive cruise control”
Researchers designed a 2nm SRAM memory cell that moves its two access transistors into…
A simulated hybrid silicon/IGO monolithic-3D 6T SRAM with back-end-of-line pass-gates gives 25% area reduction and 42% faster writes at the…
· “A Process-Aware Hybrid Si/IGO Monolithic-3D 6T SRAM with BEOL Pass-Gates for the 2nm Node”
Approximate FP multipliers cut systolic-array cost by up to 92%
Tuned per-PE arithmetic keeps neural-net accuracy while shrinking hardware by over two-thirds.
· “Precision-Aware Variable Bit Processing Elements for Hardware-Efficient Systolic Array Designs”
Auction scheduler keeps satellite downlink bidding honest
Truthful bids become the dominant strategy without sacrificing deadline guarantees.
· “STAR-GS: Truthful and Visibility-Aware Resource Scheduling for Ground Station as a Service”
To catch chip security bugs, combine five verification methods
Simulation, formal checks, lint, LLM review, and fuzzing each catch different bug classes; pre-silicon flows need them together.
· “Lessons from the Hardware Hacking Competitions: Verification Techniques, Findings, and Insights”
AI makes views cheap
Ten-year software-visualization revisit: generation is solved, perception is the bottleneck.
· “Towards Actionable Visualization: Ten Years Later, What Generative AI Changes and What It Cannot”
Do-nothing test finds a 27-swap route on a radius-10 chip
Two benchmark numbers separate uniform trust from best-route luck on tested quantum hardware.
Build agent systems around graphs
LLM agents stall on parallel, verifiable, long-horizon work; explicit task, team, and state graphs are the proposed path.
· “Graph Engineering in the Era of LLM Agents: From Individual Intelligence to System Intelligence”
QHAdamW optimizer beats Adam on Manila air quality forecasts
Combining quasi-hyperbolic momentum with decoupled weight decay lowers error and lifts AQI category accuracy.
· “Enhanced Artificial Neural Networks Using QHAdamW in Air Quality Forecasting”
Judge self-driving by safety
A survey argues learned structure, supervision, and evaluation protocol—not architecture alone—determine whether a driving policy is…
· “Planning-Oriented End-to-End Autonomous Driving: Architectures, Evaluation, and Emerging Paradigms”
Cypress cut web test time 76% and held steady over 20 runs
In a 27-case web suite, 25 tests stayed stable and data-cy locators survived UI changes with zero repair.
· “Evaluating Automated Testing on an Open-Source Web Application Using Cypress”
Multi-drone agentic AI needs oversight built in
A position paper makes interfaces, oversight, and evaluation equal partners to autonomy in safety-critical drone missions.
· “Agentic AI for Safety-critical Multi-drone Systems: Challenges and Opportunities”
NC-LID penalty lifts dynamic graph embeddings on 6 of 9 networks
A structural penalty on ambiguous boundary nodes beats plain and hub-weighted losses for reconstruction F1.
· “Enhancing Distance-Based Graph Autoencoders with Structural Penalties for Dynamic Graph Embedding”
Hybrid intelligence requires two-way adaptation
A framework defines when a living culture plus an AI becomes one co-adaptive computing system, and the benchmarks to prove it.
93% of Cessna manual pages found in top five by new retriever
Page-as-image retrieval plus vision-language answers could cut the 20-40% of technician time spent hunting manuals.
· “Reducing Technician Search Burden: A Multimodal RAG for Cessna 172 Maintenance Manual”
100 real IR frames beat 20K synthetic pretraining for drone spotting
Synthetic-first, then 100 real thermal frames: the recipe that lifts drone-detection accuracy across model families.
· “Training with synthetic data for drone detection in thermal imagery”
Phishing pages cluster into templates by HTML shape alone
Unsupervised DOM-tree fingerprints group phishing sites by construction logic, catching unnamed templates early.
· “A Tree-Structured Approach for Phishing Template and Attacker Attribution Analysis”
Open-weight agent fixes 374 of 500 SWE-bench issues
Eight independent runs cross-apply each other's tests to pick the patch, with no fine-tuning and no hidden-test peeking.
· “Kozuchi Agent: A Language-Agnostic Open-Weight Agent for Software Repair”
Energy descent generates CIFAR-10 images at FID 9.71
Class conditioning comes from a single linear field, leaving the shared interaction fabric untouched for future hardware.
Porting Chapel to RISC-V took only a handful of code changes
Benchmarks on three low-cost RISC-V boards lag server-class x86 and ARM, but the gap tracks price and power draw.
· “Porting and Benchmarking Chapel on Emerging RISC-V Hardware: an HPC Viability Study”
Federated ECG model with 37% fewer parameters beats MLP
HQKAN raises macro-F1 and minority-class recall on MIT-BIH and INCART while cutting per-round communication by up to 36%.
· “Hybrid Quantum-inspired Kolmogorov-Arnold Networks for Privacy-Aware Federated Biosignal Learning”
Humans, not bots, drove the 21x surge in AI-era pull requests
A census of 33,228 merged PRs finds more contributors and denser discussion per PR, with PR size unchanged.
A single topological cut replaces whole buffer rows in AQFP placement
Retiming-first flow cuts Josephson-junction overhead by an order of magnitude and closes timing on the full AQFP benchmark suite.
Three-stage tuning lifts C-to-Rust success from 72% to 87%
A 27B model beats much larger rivals after Rust pretraining, Verus debugging, and LeetCode task tuning.
Nine-qubit probe extracts texture order hidden in image phases
A 16-channel phase descriptor separates drug classes, malaria stages, and tissue types across three microscopy datasets.
FLSL cuts always-on leakage up to 9.8x on 22nm FDSOI
Silicon FIR and AES measurements show feedforward leakage suppression beats HVT and UHVT CMOS at low frequencies.
· “Enabling Ultra-Low-Power Always-On Feedforward Leakage Suppression Logic Circuits with FDSOI”
Inverter columns compute Gaussian splatting at 18 pJ
Threshold voltages set each Gaussian's mean and variance independently, cutting scene-reconstruction energy for edge hardware.
· “ProbSplat: Efficient Probabilistic Hardware for Gaussian Splatting in 3D Scene Reconstruction”
Trading signal length for on-device denoising cuts multi-user latency
A greedy scheduler meets each user's quality target while beating rate-only and denoise-only baselines.
Selective repair matches full recomputation in 600 decision-graph tests
With complete dependency maps it inspects only 15% of nodes; missing edges cut exact recovery to 60%.
Dithering lifts single-pixel image quality from 0.43 to 0.97 SSIM
Splitting each gray Fourier pattern into two binary ones tames noise and nears the diffraction limit.
Noise-matched gate accuracy shrinks quantum circuits by 83 percent
Layer-wise learned cost formulas replace full compilation, unlocking design-space search beyond current limits.
· “AutoQuREO: A Framework for Automated Quantum Resource Estimation and Optimization”
Classical models beat quantum for intrusion detection—mostly
A four-dataset audit finds two narrow quantum edges; the rest is classical preprocessing and regularisation.
Two graphene injectors steer a spin vector's angle
A non-local spin valve shows six resistance states as two tilted magnets reverse, matching vector addition.
· “Spin Vector Control for Heisenberg-Inspired Probabilistic Computing”
Cosine pilot reveals Doppler shifts that OTFS aliases away
A two-stage estimator pairs FFT peaks with pilot echoes, approaching perfect-CSI BER in LEO satellite channels.
· “Channel Estimation for OTFS Systems With Overspread Doppler Shifts”
4096 key candidates shrink to 13 before Grover search
Classical distinguishers prune 4096 candidates to 13, so Grover needs 2 iterations, not 50
· “Capability-Adaptive Cryptanalysis with Reduced-Space Quantum Verification”
Physical devices can now compute their own exact gradients
Reciprocity lets linear hardware train in one extra run; nonlinear trajectories need a time-reversal mirror.
Threat-aware driving policies still collapse to a constant action
Language-structured training lifts threat focus to 2.1x, yet a fixed action matches it; 12 simple policies solve 76%.
· “Language-Structured Relational Q-Learning for Threat-Aware Control in Safety-Critical Driving”
Four relay rules cut live-video frame tail from 385 ms to 114 ms
A thin rule layer between Media over QUIC and multipath QUIC is the only setup to meet the 150 ms interactive-latency budget.
· “Media-over-Multipath-QUIC for Realtime Video Applications”
Neighborly LLM agents could run 6G without a central orchestrator
Routing-style reachability ads and event-driven negotiation keep overhead near-linear while handling domain churn.
Isolation Forest flags 12 groundwater anomalies in Ghana
GRACE data from 2004–2024 yield five deficit and seven surplus months, with worst events in 2006–07.
· “Unsupervised Detection of Groundwater Storage Anomalies in Ghana Using GRACE Satellite Data”
No DNA codec wins every benchmark metric
Open-source platform scores eight codecs on density, speed, cost, and error recovery.
SVD encoding makes VQLS 10,000x faster per step
For fluid-flow linear systems, how the matrix is encoded matters more than ansatz choice; memory also drops up to 1,298x.
· “Memory-, Circuit-, and Ansatz-Efficient VQLS for CFD on Hybrid Quantum-HPC Systems”
Ring-width diversity cuts error in magnetic reservoir computer
Multiple readout channels from graded nanorings expand the usable state space, not just the parameter count.
· “Reservoir Computing with Heterogeneous Magnetic Metamaterials”
Misty-state diagrams become checkable quantum rewrites
Keeping amplitudes until the addition is done gives normalization, measurement, and equality checkable rules.
· “Certified Misty-State Rewriting (A Question-and-Answer Guide)”
One prompt beats eight agents at catching translation drift
A six-example prompt catches shifted numbers, polarities, and modalities in translated technical specs at F1=0.983.
Carbon-aware fine-tuning hits break-even in five of nine runs
One model-subject pair gains 3.5 F1 points while cutting inference CO2 by 3.5%, a strict Pareto improvement.
· “Can We Optimize the Performance-Carbon Emission Break-Even Point?: The Quest for Greener LLMs”
Trace every IoT action back to its sensor evidence
Model explanations alone cannot explain physical actions; protocols must preserve identity, time, lineage, and integrity.
LSTM chip controller prevents voltage droop
Per-chiplet voltage forecasts let the system shrink compute blocks before droop hits, keeping inference accuracy intact.
· “ReVolt: Power Delivery Network-Aware Voltage Droop Control for 2.5D PIM Chiplet Architectures”
Matrix-based SNN simulator claims up to 215x speedups on CPUs
Python package runs 100,000 spiking neurons on a laptop and claims to beat four established simulators.
· “SuperNeuroMAT: An Efficient Matrix-based Simulator for Spiking Neural Networks”
A PID layer cuts cellular-forecast drift error by 30%
The controller corrects frozen traffic models online, avoiding expensive retraining under concept drift.
· “Lightweight PID-Based Drift Mitigation for Cellular Traffic Forecasting”
Interference hurts angle estimates more than range in RIS localization
Under interference, angle error grows up to 4.4x while range error rises at most 1.3x.
· “RIS-Aided mmWave Localization Under Cross-Link Interference via Beam-Domain ML Fingerprinting”
Loss-based weighting lifts federated learning accuracy up to 7.6%
The server uses a small proxy set to down-weight high-loss clients, making aggregation stable under non-IID data and client dropout.
Ising chip plans multi-robot routes at 130x lower energy
Split into chip-sized subproblems, a 45-spin CMOS accelerator stays within 9% of a strong classical planner.
· “Ising Acceleration for Multi-Robot Multi-Target Planning”
No printed guide: AR jaw cutting guide lands within 4–7 mm
Teeth-weighted registration on HoloLens 2 tracks jaw motion in 0.8 s, pointing toward print-free mandible surgery.
DRAM behind BRAM cuts quantum circuit overhead to near zero
A control-board memory hierarchy makes deep randomized benchmarking runnable and lifts batch throughput by 43 percent.
Mortality model hits AUROC 0.852 with 605 parameters
A compact quantum LSTM plus LLM-chosen features beats a same-size classical LSTM for cardiac-arrest ICU mortality.
Same-depth qubit assignment cuts symmetry-breaking ~93 percent
A balanced placement spreads routing cost evenly at no extra depth on 108 of 124 heavy-hex patches.
Per-request simulator choice covers all 31 solvable QAOA runs
For quantum approximate optimization (QAOA) simulation, per-request pruning and ranking covers all 31 solvable runs.
· “RASP-QAOA: Resource-Aware Per-Instance Selection for Exact QAOA Simulation”
Exact QAOA gradients without the global state
Cone-local evaluation fits objective and gradient queries into a byte budget, matching a dense adjoint to 1.6e-13.
Context-gated sensing cuts wearable energy up to 75%
Viveka applies aggressive low-power sensor settings only when it trusts the activity guess, preserving accuracy within a few percent.
· “Viveka: Context-Aware Sensing for Energy Efficiency in Smart Wearables”
A qubit per amino acid samples protein energy landscapes
Repeated measurement yields residue-level microstates, funnel-like energy distributions, and key-residue maps.
· “A Quantum Circuit Framework for Protein Ensemble-Level Energetics”
Prune files first, then run per-file ANN: 32x faster filtered search
The table's own metadata does the filtering, so vectors stay in one open format with no copied-out vector store.
Circuit cutting merges gate, wire, and joint cuts
Graph partitioning finds near-optimal cut placements for hundreds of qubits, with up to 90% lower sampling overhead
· “Scalable Circuit Cutting: A Framework for Combined Gate and Wire Cuts Using Gate Groups”
Pose-only student beats larger models on classroom incidents
Distilled from multi-order motion reasoning, it hits 71.78% on synthetic and 63.41% on real CCTV footage.
· “Robust and Efficient Motion Reasoning for Privacy-Aware Classroom Incident Recognition”
Blockage-aware bandit cuts 6G-V2X regret by 59 percent
New algorithms share learned mmWave link knowledge between vehicles, cutting cold start and false alarms.
Shallow QAOA landscapes grow rougher
Across 23,000 MIS instances, variance grows with system size, undercutting Lie-algebra barren-plateau predictions.
Static power can flip the PIM-vs-GPU efficiency verdict
Idle DRAM and GPU power dominate LLM decode efficiency, so dynamic-only PIM comparisons mislead architects.
· “On Design Principles for Efficient Heterogeneous DRAM-PIM-GPU Systems”
Tool exposes why a 'relaxed' routing shortcut backfired
NAViz, an open-source app for neutral-atom quantum computers, spotted the detours and inspired a faster hybrid strategy.
· “Guiding Compiler Optimizations for Neutral Atom Quantum Computers Through Visualizations”
Personalized monitoring flags cognitive decline years before diagnosis
Comparing scores with a person's own trajectory, not population norms, catches decline up to 6 years early.
University AI policies guard data
Nationwide scan of 130 US universities finds AI rules split by level: schools teach tools, universities manage risk.
Couple two worlds: Enactive AI grounds decisions in real systems
A four-part architecture ties purpose, constraints, execution, and feedback into one decision cycle
· “Enactive Artificial Intelligence: A Decision-Centric Architecture for Complex Systems”
Schedule-aware model cuts checkpoint forecast error to 9.33%
Pre-departure arrival signals let a Temporal Fusion Transformer outperform RNN and LSTM baselines on six-hour forecasts.
· “Schedule-Informed Temporal Fusion Forecasting of Hourly Airport Security-Checkpoint Throughput”