MIT researchers have demonatrated a new architecture for an ultranarrow linewidth integrated laser based on stabilization to a sinusoidal fringe of an interferometer having a long 25 meters unbalanced delay line. arxiv.org/pdf/2602.16461

Reseachers have demonstrated quantum teleportation over Deutsche Telekom’s metropolitan fiber testbed in Berlin using commercial components deployed at the telecom datacenter. arxiv.org/pdf/2602.16613

Researchers have demonstrated real-time intermodal free-space QKD at telecommunication wavelength over an 18 km ground-to-ground low-altitude link, effectively exploiting adaptive optics in a 40 cm-class telescope to mitigate atmospheric turbulence. arxiv.org/pdf/2602.16680

In this paper is presented DARTH-PUM, a general-purpose hybrid processing-using-memory (PUM) architecture which combines both analog and digital PUM onto the same chip. arxiv.org/pdf/2602.16075

In this paper is presented ROIX-Comp, a region-of-interest driven extraction framework that intelligently compresses X-ray computed tomography data by identifying and retaining only essential features. arxiv.org/pdf/2602.15917

In this paper, researchers have proposed a design of a new GPU communication API that can support CPU-free communication and mostly preserves familiar MPI two-sided communication abstractions. arxiv.org/pdf/2602.15356

In this paper is introduced FlashMem, a memory streaming framework designed to efficiently execute large-scale modern DNNs and multi-DNN workloads while minimizing memory consumption and reducing inference latency. arxiv.org/pdf/2602.15379

In this paper, researches have demonstrated a robust framework for generating and controlling fractional optical skyrmions mediated by fractional OAM modes, paving the way for high-density optical information encoding. arxiv.org/pdf/2602.15464

In this paper is introduced PRISM, a photonics-informed neural inverse lithography framework that turns inverse-designed photonic integrated circuits layouts into high-yield, manufacturable hardware. arxiv.org/pdf/2602.15762

In this paper is presented Scope, a novel merged pipeline framework for MCM-based NN inference that enhances hardware utilization and reduces network-on-package overheads, while also mitigating segment count tradeoffs in segmented pipelines. arxiv.org/pdf/2602.14393

In this paper is presented GLM-5, a foundation model designed to transition the paradigm from vibe coding to agentic engineering, adopting DeepSeek Sparse Attention to reduce training and inference costs while maintaining long-context fidelity. arxiv.org/pdf/2602.15763

In this paper is presented OptiML, an end-to-end generate-then-optimize framework that unifies a Mixture-of-Thoughts proposal policy generator with search-based optimization via Monte Carlo Tree Search. arxiv.org/pdf/2602.12305

In this paper is introduced Neural Irradiance Volume, a neural-based technique that allows accurate real-time rendering of diffuse global illumination via a compact pre-computed model, overcoming the limitations of traditional probe-based methods. arxiv.org/pdf/2602.12949

Samsung researchers have proposed TriGen, a novel NPU architecture for end- to-end acceleration of LLM inferences based on SW-HW co-design, which adopts MXINT8 number system and HW resource constraint aware scheduling. arxiv.org/pdf/2602.12962

In this paper is presented MXFormer, a novel hybrid weight-stationary CiM accelerator that provides high throughput and efficiency for fixed-model inference on large short-sequence Transformers. arxiv.org/pdf/2602.12480

In this paper is presented DPUConfig, a custom RL- based agent that continuously learns and selects near-optimal DPU configurations by taking into account the features of the ML model and dynamically adapting to the stochastic runtime variance. arxiv.org/pdf/2602.12847

In this lecture notes is presented the theory of learning in deep and recurrent NNs from the PoV of Bayesian inference, providing an introduction into some of the useful techniques from statistical physics, disordered systems, and large deviation theory. arxiv.org/pdf/2602.12855

In this paper is proposed Gabor Fields, a new volumetric primitive fields that enables continuous frequency filtering at no cost, enabling continuous LoD control, reduced rendering time, and a trade-off between variance and bias for performance. arxiv.org/pdf/2602.05081

In this paper is presented CVA6-CFI, the first design, integration, and evaluation of the standard #RISC-V extensions for Control-Flow Integrity, introducing 2 independent hardware units implementing backward-edge and forward-edge control-flow protection. arxiv.org/pdf/2602.04991

In this paper, researchers have analyzed the neural scaling behavior for a high-dimensional spectroscopic learning task using over 200k ab initio dielectric functions, finding the emergence of broken neural scaling in the field of materials science. arxiv.org/pdf/2602.05702

In this paper, researchers have demonstrated the coexistence of room-temperature out-of-plane ferroelectricity and metallic conductivity with ultrahigh carrier density in atomically thin TaNiTe5. arxiv.org/pdf/2602.05398

In this paper is presented an empirical study of LDPC 5G decoding on an NVIDIA DGX Spark, comparing execution on the Grace CPU and the integrated Blackwell GB10 GPU using a Sionna PHY/SYS implementation of a 5G NR–like link-level chain. arxiv.org/pdf/2602.04652

In this paper is presented Crypto-RV, a RISC-V coprocessor architecture that unifies support for SHA-256, SHA-512, SM3, SHA3-256, SHAKE-128, SHAKE-256 AES-128, HARAKA-256, and HARAKA-512 within a single 64-bit datapath. #RISCV arxiv.org/pdf/2602.04415

In this paper, researchers have demonstrated that Pna21/ CaSnN2 is a potential alternative to InxGa1−xN in blue LEDs and for the technology of adding a yellow phosphorescent coating for white LEDs. arxiv.org/pdf/2602.04106

Researchers have demonstrated a dual-modulated vertical transistor with a leakage-blocking layer, fabricated through a simple stacked process without the need for high-resolution alignment or costly techniques, even at nanoscale channel lengths. advanced.onlinelibrary.wiley.com/doi/epdf/10....

In this thread, I will highlight some of the works that will be presented in February at ISSCC 2026. The following thread highlights some of the most interesting works of this year. There are a lot of interesting news on the way... Advance Program: submissions.mirasmart.com/ISSCC2026/PD...

In this paper is proposed HetCCL, the first cross-vendor collective communication library that enables DL model training and inference on heterogeneous clusters with both NVIDIA and AMD GPUs. arxiv.org/pdf/2601.22585

Researchers have presented a comprehensive analysis of the tight-binding properties of twisted and strained bilayer graphene, revealing a strong dependence of band narrowing and topology on strain type, magnitude, direction and lattice relaxation. arxiv.org/pdf/2602.02692

In this paper is presented a rigorous error analysis of the Ozaki-II scheme, showing the accuracy behavior of the method and enabling the estimation of the number of low-precision matrix mult. required to achieve a desired level of numerical accuracy. #HPC arxiv.org/pdf/2602.02549

In this paper is presented Phoenix, a modular pointer analysis framework for C/C++ that unifies multiple state-of-the-art alias analysis algorithms behind a single, stable interface. arxiv.org/pdf/2602.01720

I would like to invite all my followers to subscribe on my Substack. Everyone who loves technology and is hungry for new discoveries is welcome. Also, feel free to contact me! Profile Link: underfox3.substack.com

In this paper is proposed HetCCL, the first cross-vendor collective communication library that enables DL model training and inference on heterogeneous clusters with both NVIDIA and AMD GPUs. arxiv.org/pdf/2601.22585

In this paper is proposed KHEPRI, a heterogeneous GPU architecture for rendering graphics applications that assigns compute-intensive tiles to compute-specialized cores and memory-intensive tiles to memory-specialized cores. arxiv.org/pdf/2601.22862

Musk's main idea of ​​creating orbital data centers may be difficult, but it's not technologically impossible to achieve. However, this will require hardware appropriate to the space environment, which both Altera and AMD are already prepared to provide. www.spacex.com/updates#xai-...

In this paper is presented AscendCraft, a DSL-guided framework for automatic AscendC kernel generation using LLM which introduces a lightweight DSL that abstracts non-essential complexity while explicitly modeling Ascend-specific execution semantics. #HPC arxiv.org/pdf/2601.22760

Nvidia's $100 billion investment contract in OpenAI could easily be considered an antitrust violation if we analyze it in light of the contractual terms and their consequences for the market... I hope Jensen is wearing his leather jacket when he's called to court.

In this thread, I will highlight some of the works that will be presented in February at ISSCC 2026. The following thread highlights some of the most interesting works of this year. There are a lot of interesting news on the way... Advance Program: submissions.mirasmart.com/ISSCC2026/PD...

Researchers have investigated the channel width scaling in nanoribbon transistors based on monolayer MoS2 grown on 2-inch wafers, revealing excellent electrostatic control, minimal short-channel effects, and low device-to-device variability. arxiv.org/pdf/2601.13696

In this paper, researchers have demostrated that acyclic deterministic finite automatons, as well as Moore and Mealy machines, from prefix-complete instances in a binary alphabet, when the instance strings are all prefixes of a single string, is NP-complete. arxiv.org/pdf/2601.12621

Researchers have proposed Jet-RL, an on-policy FP8 reinforcement learning framework that enforces a unified precision workflow for training and deployment, resolving policy mismatches and ensuring stable optimization. arxiv.org/pdf/2601.14243

In this paper is presented DaggerFFT, a distributed FFT framework in Julia that treats highly parallel FFT computations as a dynamically scheduled task graph with each FFT stage operates on a separately defined distributed array. arxiv.org/pdf/2601.12209

In this paper is presented a detailed analysis focused on quantifying and improving #HPC resource utilization efficiency specifically within the Latin American Giant Observatory computational environment. arxiv.org/pdf/2601.13424

In this paper is proposed OPTIMUM-DERAM, a novel decentralized shared memory framework for Web3 that leverages Random Linear Network Coding to enable atomic read/write operation. arxiv.org/pdf/2601.13146

In this paper is presented Memory Disorder: a timerless side-channel that uses memory re-orderings to infer activity on other processes. arxiv.org/pdf/2601.08770

In this paper is presented a high-performance, mixed-precision solver for symmetric linear systems based on a nested recursive Cholesky decomposition that uniformly applies recursion across POTRF, TRSM, and SYRK phases. arxiv.org/pdf/2601.08082

In this paper is presented MatrixPIC, the first holistic co-design of the Particle-in-Cell deposition kernel, data layout, and an incremental sorting mechanism, all tailored specifically for the hybrid MPU-VPU SIMD execution model on modern CPUs. arxiv.org/pdf/2601.08277

Google researchers have proposed a single-stage Huffman encoder for machine learning compression which use fixed codebooks derived from the average probability distribution of previous data batches, reducing computational, latency and data overheads. arxiv.org/pdf/2601.10673

In this paper, researchers have demonstrated a fully on-chip probabilistic computer for dense combinatorial optimization, using MIMO detection as a benchmark. arxiv.org/pdf/2601.09037

In this paper is presented a systematic evaluation of NVIDIA’s Blackwell consumer GPUs for production LLM inference, benchmarking four open-weight models across 79 configurations spanning quantization formats, context lengths, and three workloads. arxiv.org/pdf/2601.09527

In this paper is presented DPA-Store, an ordered in-memory KV store that uses the BlueField-3 SmartNIC and its data path accelerators to bypass OS overheads and to allow stateless clients. arxiv.org/pdf/2601.06231