AI Factory Glossary

u chart, spc

Defects per unit.

u-net denoiser, generative models

Architecture for diffusion models.

u-shaped cell, manufacturing operations

U-shaped cells position equipment in U-configuration enabling operator flexibility and communication.

ucie, business & strategy

Universal Chiplet Interconnect Express provides standard interface for chiplet communication.

uda, uda, semi-supervised learning

Consistency with strong augmentation.

ulmfit, transfer learning

Transfer learning method for NLP.

ulpa filter (ultra-low particulate air),ulpa filter,ultra-low particulate air,facility

Even more efficient than HEPA removes 99.999% of particles.

ultimate sd upscale, generative models

Tiled upscaling method.

ultra-low-k (ulk),ultra-low-k,ulk,beol

k < 2.5 dielectrics.

ultra-thin body and box (utbb),ultra-thin body and box,utbb,technology

Advanced FD-SOI with thin BOX.

ultralytics,yolo,detection

Ultralytics provides YOLOv8 and HUB. Easy object detection. Training and export.

ultrasonic testing, manufacturing operations

Ultrasonic testing detects internal flaws leaks or thickness changes through sound wave analysis.

ultraviolet photoelectron spectroscopy, ups, metrology

Measure valence band and work function.

umbrella sampling, chemistry ai

Sample along reaction coordinate.

unbiased hast, reliability

HAST without electrical bias.

unbiased learning rank, recommendation systems

Unbiased learning to rank estimates relevance from biased click data using inverse propensity scoring or doubly robust estimators.

uncertainty budget, metrology

Breakdown of uncertainty sources.

uncertainty quantification, ai safety

Uncertainty quantification estimates model confidence in predictions.

uncertainty quantification,ai safety

Estimate confidence in predictions.

uncertainty-based rejection,ai safety

Abstain based on uncertainty estimates.

uncertainty,confidence,epistemic

Uncertainty quantification distinguishes epistemic (model) from aleatoric (data) uncertainty. Know when to abstain.

under-sampling majority class, machine learning

Remove common examples.

underfill filler, packaging

Particles in underfill.

underfill for cte matching, advanced packaging

Material reducing CTE mismatch effects.

underfill for tsv, advanced packaging

Material filling gaps around TSV.

underfill process, packaging

Fill gap under flip-chip.

underfill voids, packaging

Trapped air in underfill.

underfill,advanced packaging

Epoxy material filling gap between die and substrate for reliability.

undersampling,balance,reduce

Undersampling removes majority class examples. Balance classes.

undershoot, signal & power integrity

Undershoot is signal excursion below ground or supply during transitions potentially causing latchup.

undertraining, training

Insufficient training.

unicode normalization, nlp

Standardize Unicode representations.

unidirectional attention,transformer

Each token only attends to previous tokens (GPT-style).

unified memory, infrastructure

Shared memory space CPU-GPU.

unified vision-language models,multimodal ai

Models handling multiple vision-language tasks.

uniformity (cvd),uniformity,cvd

Thickness variation across wafer.

unigram tokenization, nlp

Probabilistic tokenization.

unipc sampling, generative models

Unified predictor-corrector sampler.

universal adversarial triggers,ai safety

Inputs that reliably cause unwanted behavior.

universal adversarial, interpretability

Universal adversarial perturbations cause misclassification across multiple inputs with single perturbation.

universal chiplet interconnect express, standards

Standard for chiplet interconnection.

universal domain adaptation, domain adaptation

Handle target classes not in source.

universal transformers,llm architecture

Recurrent depth with adaptive computation.

universal value function approximators, uvfa, reinforcement learning

Generalize across goals.

universal value function, reinforcement learning advanced

Universal value functions generalize across goals and states simultaneously enabling transfer and multi-task learning.

universally slimmable networks, neural architecture

Switch between widths seamlessly.

univnet, audio & speech

Universal Neural Vocoder uses multi-resolution spectrogram discriminators for high-fidelity speech synthesis.

unk token, unk, nlp

Unknown/out-of-vocabulary token.

unlearning,ai safety

Remove specific knowledge or capabilities from a trained model.

unlearning,forget,remove

Machine unlearning removes specific knowledge from trained models. Privacy, copyright, safety applications.