英特尔公司

Accurate measurements of the nanometer scale geometry of extreme ultraviolet (EUV) lithography photoresist (PR) patterns are a critical metrology step in semiconductor manufacturing. The atomic force microscope (AFM) can nondestructively measure three-dimensional surface PR profiles with nanometer resolution. However, geometric and mechanical artifacts limit AFM's accuracy when measuring tall or narrow features comparable in size to the AFM tip radius. In this work, we investigate tip-sample interaction dynamics on 40 nm pitch EUV PR features using force mapping-based AFM with high-aspect ratio diamond-like carbon spike tips. We train and utilize a random forest machine learning algorithm to classify force curves based on their linearity, adhesion, and hysteresis. Some of the observed force curves are consistent with stick-slip friction between the AFM tip and PR sidewall during indentation. We compare the AFM measured PR profile to those measured with scanning electron microscopy. We conclude that a combination of PR shrinkage under electron beam exposure, geometric dilation of the profile due to the finite size of the AFM tip, and bending of the AFM tip as it interacts with the sidewall can explain the experimental results. This study challenges the conventional assumption that high aspect ratio tips with smaller tip radius have capability to reach narrow features; instead, we demonstrate that complex tip-sample interaction dynamics occur around the sidewall. This work contributes to an improved understanding of AFM measurement artifacts associated with high-aspect ratio spike tips on nanometer scale patterns.

Inference providers must maintain availability for many LLMs, including low-volume but essential models, making resource efficiency increasingly important as token prices fall. Analysis of production traces reveals a dynamic bursty-group pattern in which sets of models become active together and shift over time; existing space- and time-sharing approaches lack principled mechanisms to adapt to this variability, forcing trade-offs between SLO adherence and efficiency. We observe that elastic memory allocation can unify spatial and temporal sharing. Based on this insight, we have developed Prism, a memory-centric LLM co-serving framework that applies memory ballooning to reclaim memory across models and support both forms of sharing under a single scheme. Prism's balloon driver, referred to as kvcached, has been open-sourced at https://github.com/ovg-project/kvcached, and deployed in production environments across 10K+ GPUs.

In this paper, we investigate the degradation of the thermal conductivity of diamond due to the presence of the isotope C-13. We employ a full-band Monte Carlo simulation in which phonons are modeled as semiclassical particles, and the full nonlinear phonon Boltzmann transport equation is solved. The thermal conductivity of diamond with C-13 concentrations of 0.005% (isotopically ultrapure), 0.1% (isotopically enriched), and 1.07% (naturally occurring) is computed for a temperature range of 80-500 K, showing good agreement with experimental data. The degradation of thermal conductivity with increasing isotope concentration is observed across the entire temperature range and is significant at low temperatures. At 80 K, the thermal conductivity of the naturally occurring sample is about four times lower than that of the isotopically enriched sample and eleven times lower than that of the isotopically ultrapure one. We further investigate the isotope effect on individual phonon modes and compute their individual contribution to the thermal conductivity. We also calculate the expected value of the mean free path and relaxation time of all phonon modes' population. Acoustic phonons are the principal contributors to thermal conductivity across all isotope concentrations. The optical phonon modes present relaxation times and mean free paths that are affected by the isotope concentration but are fairly constant over the entire temperature range. The acoustic phonon mean free paths decrease with increasing isotope concentration and are temperature-dependent, ranging from 2 mm to 0.5 mu m.

Numerous applications of large language models (LLMs) rely on their ability to perform step-by-step reasoning. However, the reasoning behavior of LLMs remains poorly understood, posing challenges to research, development, and safety. To address this gap, we introduce landscape of thoughts (LoT), the first landscape visualization tool to inspect the reasoning trajectories with certain reasoning methods on any multi-choice dataset. We represent the textual states in a trajectory as numerical features that quantify the states' distances to the answer choices. These features are then visualized in two-dimensional plots using t-SNE. Qualitative and quantitative analysis with the landscape of thoughts effectively distinguishes between strong and weak models, correct and incorrect answers, as well as different reasoning tasks. It also uncovers undesirable reasoning patterns, such as low consistency and high uncertainty. Additionally, users can adapt LoT to a model that predicts the property they observe. We showcase this advantage by adapting LoT to a lightweight verifier that evaluates the correctness of trajectories. Empirically, this verifier boosts the reasoning accuracy and the test-time scaling effect. The code is publicly available at: https://github.com/tmlr-group/landscape-of-thoughts.

Large language models (LLMs) are helping millions of users write texts about diverse issues, and in doing so expose users to different ideas and perspectives. This creates concerns about issue bias, where an LLM tends to present just one perspective on a given issue, which in turn may influence how users think about this issue. So far, it has not been possible to measure which issue biases LLMs actually manifest in real user interactions, making it difficult to address the risks from biased LLMs. Therefore, we create IssueBench: a set of 2.49m realistic prompts for measuring issue bias in LLM writing assistance, which we construct based on 3.9k templates (e.g. "write a blog about") and 212 political issues (e.g. "AI regulation") from real user interactions. Using IssueBench, we show that issue biases are common and persistent in state-of-the-art LLMs. We also show that biases are remarkably similar across models, and that all models align more with US Democrat than Republican voter opinion on a subset of issues. IssueBench can easily be adapted to include other issues, templates, or tasks. By enabling robust and realistic measurement, we hope that IssueBench can bring a new quality of evidence to ongoing discussions about LLM biases and how to address them.