arXiv subjects

Yuji Yamamoto

Publications and source records attributed to Yuji Yamamoto.

14 recordsLinked to original sources

arXiv2026

It's Not RoPE that Creates Sinks: The Role of Self-Concentration and Value-Non-Mixing in Attention

Large Language Models (LLMs) often exhibit "Attention Sink" (AS) and the accompanying "Massive Activations" (MAs) at the initial position of a sequence. These phenomena frequently co-occur, and MAs can pose challenges for low-bit quantization. In this study, we analyze the factors underlying AS and MAs that emerge at the initial position regardless of the token occupying it. Our experiments suggest that self-concentration of attention, resulting from the causal mask, and the subsequent Value-non-mixing in attention outputs contribute to AS and MAs. These findings provide new empirical evidence on the internal dynamics of LLMs, offering insights that may inform future quantization strategies and advance our understanding of the internal mechanisms of attention layers.

cs.CL↗

arXiv2026

Mobility Enhancement in Si/SiGe Quantum Well Enabled by a Buried Si Layer Trapping Oxygen Impurities

Reducing disorder in undoped Si/SiGe field-effect heterostructures remains an important materials challenge for scalable quantum devices, particularly electron spin qubits. Background impurities such as oxygen have been identified as mobility-limiting, yet practical heterostructure-design strategies for suppressing their incorporation remain underexplored, and their influence across different transport regimes is not fully established. Here, we demonstrate a simple route to oxygen reduction and mobility enhancement in Si/SiGe quantum-well heterostructures grown by reduced-pressure chemical vapor deposition (RP-CVD) on 200 mm Si(100) substrates through the introduction of a thin, electrically passive buried Si layer within the lower SiGe barrier. Secondary-ion mass spectrometry shows that the buried Si layer reproducibly reduces the oxygen background in the subsequently grown SiGe by approximately a factor of five, without modifying the active quantum-well region. Density- and temperature-dependent magnetotransport measurements further show that this reduction increases the electron mobility, while leaving the percolation density and density-dependent mobility scaling largely unchanged. Upon cooling to 0.3 K, both high- and low-oxygen devices exhibit similar density-dependent fractional mobility enhancements, indicating that the reduced oxygen background improves momentum relaxation without substantially altering the dominant low-density disorder landscape. These results establish the buried Si layer as a straightforward and process-compatible heterostructure-design element for reducing oxygen incorporation and improving transport in 200 mm CVD-grown Si/SiGe quantum-device materials.

cond-mat.mes-hall↗

arXiv2026

Disorder signatures emerging at millikelvin temperatures in Si/SiGe field-effect stacks

The performance and scalability of electron spin qubits based on gate-defined quantum-dots in undoped Si/SiGe field-effect stacks remain constrained by disorder originating from the gate stack. Its coupling to the quantum well can be reduced by increasing the Si quantum well depth, while electrostatic charge history, for example through interface-trap filling, can further modify the effective disorder landscape. Although such disorder is commonly benchmarked through mobility measurements using magnetotransport and Hall bar devices, dedicated investigations at millikelvin temperatures relevant for quantum-dot operation remain limited. Here, we use temperature-dependent magnetotransport on Hall bar shaped field-effect transistors to investigate how mobility-based disorder signatures depend on quantum-well depth and charge history from $1.5~\mathrm{K}$ down to the millikelvin regime. We show that magnetotransport characterization at $1.5~\mathrm{K}$ captures the dominant mobility improvement associated with reduced dielectric-interface coupling, but can underestimate disorder differences that emerge at millikelvin temperatures, particularly in the low-density regime. Our results therefore highlight that millikelvin magnetotransport characterization of Hall bar devices can provide additional insight for optimizing Si/SiGe field-effect stacks, particularly in the context of gate-defined quantum dot spin qubits.

cond-mat.mes-hall↗

arXiv2026

Timesteps of Mamba Align with Human Reading Times

This study demonstrates an alignment of per-word processing time in a popular state-space language model Mamba and human readers. In Mamba, the recurrent state transition at each layer conceptually takes some duration of time, the discretization timestep $\Delta_t$, determined dynamically in response to the input. Using a naturalistic reading dataset, we show that the per-word timestep from Mamba is a significant predictor of human reading times, and remains significant even when known predictors such as GPT-2 surprisal are controlled for. We further suggest, through formal analysis of Mamba's architecture and internal dynamics, that Mamba can serve as a new, valuable lens to look at human real-time language processing with ever-updated memory, because it allows us to look at how each module (layer) weighs short- and long-term information retention, and how noise may interact with dynamic, continuous memory representation. Code is available online.

cs.CL↗

arXiv2026

Bit-Flip Vulnerability of Shared KV-Cache Blocks in LLM Serving Systems

Rowhammer on GPU DRAM has enabled adversarial bit flips in model weights; shared KV-cache blocks in LLM serving systems present an analogous but previously unexamined target. In vLLM's Prefix Caching, these blocks exist as a single physical copy without integrity protection. Using software fault injection under ideal bit targeting, we characterize worst-case severity and identify three properties: (1) Silent divergence - 13 of 16 BF16 bit positions produce coherent but altered outputs, indistinguishable from legitimate responses without a clean baseline. (2) Selective propagation - only requests sharing the targeted prefix are affected. (3) Persistent accumulation - no temporal decay occurs, so cumulative damage grows linearly with subsequent requests. Together, these constitute a threat profile distinct from weight corruption: silent divergence and selective propagation enable detection evasion; persistent accumulation then proceeds unchecked, yielding damage amplification bounded only by how long the block remains cached. A checksum-based countermeasure detects any single-bit corruption at scheduling time, bounding cumulative damage to one batch independent of the block's cache lifetime, with negligible overhead. These results argue for integrity protection of prefix blocks before end-to-end exploitation is demonstrated.

cs.CR↗

arXiv2025

High yield, low disorder Si/SiGe heterostructures for spin qubit devices manufactured in a BiCMOS pilot line

The prospect of achieving fault-tolerant quantum computing with semiconductor spin qubits in Si/SiGe heterostructures relies on the integration of a large number of identical devices, a feat achievable through a scalable (Bi)CMOS manufacturing approach. To this end, both the gate stack and the Si/SiGe heterostructure must be of high quality, exhibiting uniformity across the wafer and consistent performance across multiple fabrication runs. Here, we report a comprehensive investigation of Si/SiGe heterostructures and gate stacks, fabricated in an industry-standard 200 mm BiCMOS pilot line. We evaluate the homogeneity and reproducibility by probing the properties of the two-dimensional electron gas (2DEG) in the shallow silicon quantum well through magnetotransport characterization of Hall bar-shaped field-effect transistors at 1.5 K. Across all the probed wafers, we observe minimal variation of the 2DEG properties, with an average maximum mobility of $(4.25\pm0.17)\times 10^{5}$ cm$^{2}$/Vs and low percolation carrier density of $(5.9\pm0.18)\times 10^{10}$ cm$^{-2}$ evidencing low disorder potential in the quantum well. The observed narrow statistical distribution of the transport properties highlights the reproducibility and the stability of the fabrication process. Furthermore, wafer-scale characterization of a selected individual wafer evidenced the homogeneity of the device performances across the wafer area. Based on these findings, we conclude that our material and processes provide a suitable platform for the development of scalable, Si/SiGe-based quantum devices.

cond-mat.mes-hall↗

arXiv2024

Industrially fabricated single-electron quantum dots in Si/Si-Ge heterostructures

This paper reports the compatibility of heterostructure-based spin qubit devices with industrial CMOS technology. It features Si/Si-Ge quantum dot devices fabricated using Infineon's 200 mm production line within a restricted thermal budget. The devices exhibit state-of-the-art charge sensing, charge noise and valley splitting characteristics, showing that industrial fabrication is not harming the heterostructure quality. These measured parameters are all correlated to spin qubit coherence and qubit gate fidelity. We describe the single electron device layout, design and its fabrication process using electron beam lithography. The incorporated standard 90 nm back-end of line flow for gate-layer independent contacting and wiring can be scaled up to multiple wiring layers for scalable quantum computing architectures. In addition, we present millikelvin characterization results. Our work exemplifies the potential of industrial fabrication methods to harness the inherent CMOS-compatibility of the Si/Si-Ge material system, despite being restricted to a reduced thermal budget. It paves the way for advanced quantum processor architectures with high yield and device quality.

cond-mat.mes-hall↗

arXiv2024

Impact of biased cooling on the operation of undoped silicon quantum well field-effect devices for quantum circuit applications

Gate-tunable semiconductor nanosystems are getting more and more important in the realization of quantum circuits. While such devices are typically cooled to operation temperature with zero bias applied to the gate, biased cooling corresponds to a non-zero gate voltage being applied before reaching the operation temperature. We systematically study the effect of biased cooling on different undoped SiGe/Si/SiGe quantum well field-effect stacks (FESs), designed to accumulate and density-tune two-dimensional electron gases (2DEGs). In an empirical model, we show that biased cooling of the undoped FES induces a static electric field, which is constant at operation temperature and superimposes onto the field exerted by the top gate onto the 2DEG. We show that the voltage operation window of the field-effect-tuned 2DEG can be chosen in a wide range of voltages via the choice of the biased cooling voltage. Importantly, quality features of the 2DEG such as the mobility or the temporal stability of the 2DEG density remain unaltered under biased cooling. We discuss how this additional degree of freedom in the tunability of FESs may be relevant for the operation of quantum circuits, in particular for the electrostatic control of spin qubits.

cond-mat.mes-hall↗

arXiv2023

Local laser-induced solid-phase recrystallization of phosphorus-implanted Si/SiGe heterostructures for contacts below 4.2 K

Si/SiGe heterostructures are of high interest for high mobility transistor and qubit applications, specifically for operations below 4.2 K. In order to optimize parameters such as charge mobility, built-in strain, electrostatic disorder, charge noise and valley splitting, these heterostructures require Ge concentration profiles close to mono-layer precision. Ohmic contacts to undoped heterostructures are usually facilitated by a global annealing step activating implanted dopants, but compromising the carefully engineered layer stack due to atom diffusion and strain relaxation in the active device region. We demonstrate a local laser-based annealing process for recrystallization of ion-implanted contacts in SiGe, greatly reducing the thermal load on the active device area. To quickly adapt this process to the constantly evolving heterostructures, we deploy a calibration procedure based exclusively on optical inspection at room-temperature. We measure the electron mobility and contact resistance of laser annealed Hall bars at temperatures below 4.2 K and obtain values similar or superior than that of a globally annealed reference samples. This highlights the usefulness of laser-based annealing to take full advantage of high-performance Si/SiGe heterostructures.

physics.app-ph↗

arXiv2019

Weak-coupling Mean-field Theory of Magnetic Properties of NiGa$_2$S$_4$

We report a mean-field theoretical study of a triangular lattice magnet NiGa$_2$S$_4$. Specifically, spiral mean-field theory is applied to a 17-band $d$-$p$ model constructed from the maximally localized Wannier functions. Our itinerant-model approach shows that the most stable spiral magnetic state has an ordering vector near ${\bf Q}=(0.15,0.15,0)$, consistent with neutron scattering experiments, when we assume the Ni-site Coulomb interaction is not so strong ($U \approx 2$ eV). To map onto a classical Heisenberg spin model, we estimate spin exchange interactions from the mean-field results, and find that the nearest-neighbor exchange is ferromagnetic and the largest (larger than the third nearest-neighbor exchange, in contrast to early studies). We also calculate the dynamical spin correlation function $S({\bf q}, \omega)$, using the same model within the random-phase approximation (RPA). Calculated $S({\bf q}, \omega)$ has a spectral structure quite different from that of conventional spin-wave excitations.

cond-mat.str-el↗

arXiv2019

Strain and Band-Gap Engineering in Ge-Sn Alloys via P Doping

Ge with a quasi-direct band gap can be realized by strain engineering, alloying with Sn, or ultrahigh n-type doping. In this work, we use all three approaches together to fabricate direct-band-gap Ge-Sn alloys. The heavily doped n-type Ge-Sn is realized with CMOS-compatible nonequilibrium material processing. P is used to form highly doped n-type Ge-Sn layers and to modify the lattice parameter of P-doped Ge-Sn alloys. The strain engineering in heavily-P-doped Ge-Sn films is confirmed by x-ray diffraction and micro Raman spectroscopy. The change of the band gap in P-doped Ge-Sn alloy as a function of P concentration is theoretically predicted by density functional theory and experimentally verified by near-infrared spectroscopic ellipsometry. According to the shift of the absorption edge, it is shown that for an electron concentration greater than 1x10^20 cm-3 the band-gap renormalization is partially compensated by the Burstein-Moss effect. These results indicate that Ge-based materials have high potential for use in near-infrared optoelectronic devices, fully compatible with CMOS technology.

cond-mat.mtrl-sci↗

arXiv2014

Emergence of self-similarity in football dynamics

The multiplayer dynamics of a football game is analyzed to unveil self-similarities in the time evolution of player and ball positioning. Temporal fluctuations in both the team-turf boundary and the ball location are uncovered to follow the rules of fractional Brownian motion with a Hurst exponent of H=0.7. The persistence time below which self-similarity holds is found to be several tens of seconds, implying a characteristic time scale that governs far-from-equilibrium motion on a playing field.

physics.pop-ph↗

arXiv2013

Graphene Grown on Ge(001) from Atomic Source

Among the many anticipated applications of graphene, some - such as transistors for Si microelectronics - would greatly benefit from the possibility to deposit graphene directly on a semiconductor grown on a Si wafer. We report that Ge(001) layers on Si(001) wafers can be uniformly covered with graphene at temperatures between 800°C and the melting temperature of Ge. The graphene is closed, with sheet resistivity strongly decreasing with growth temperature, weakly decreasing with the amount of deposited C, and reaching down to 2 kOhm/sq. Activation energy of surface roughness is low (about 0.66 eV) and constant throughout the range of temperatures in which graphene is formed. Density functional theory calculations indicate that the major physical processes affecting the growth are: (1) substitution of Ge in surface dimers by C, (2) interaction between C clusters and Ge monomers, and (3) formation of chemical bonds between graphene edge and Ge(001), and that the processes 1 and 2 are surpassed by CH$_{2}$ surface diffusion when the C atoms are delivered from CH$_{4}$. The results of this study indicate that graphene can be produced directly at the active region of the transistor in a process compatible with the Si technology.

cond-mat.mes-hall↗

arXiv2012

Molecular beam growth of graphene on mica

We demonstrate molecular beam growth of graphene on biotite mica substrates at temperatures below 1000°C. As indicated by optical and atomic force microscopy, evaporation of carbon from a high purity solid-state source onto biotite surface results in the formation of single-, bi-, and multilayer graphene with size in the micrometer regime. Graphene grown directly on mica surface is of very high crystalline quality with the defect density below the threshold detectable by Raman spectroscopy. The interaction between graphene and the mica substrate is studied by comparison of the Raman spectroscopy and atomic force microscopy data with the corresponding results obtained for graphene flakes mechanically exfoliated onto biotite substrates. Experimental insights are combined with density functional theory calculations to propose a model for the initial stage of the van der Waals growth of graphene on mica surfaces. This work provides important hints on how the direct growth of high quality graphene on insulators can be realized in general without exceeding the thermal budget limitations of Si technologies.

cond-mat.mes-hall↗