AI Factory Glossary

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.

unobserved components, time series models

Unobserved components models represent time series as sums of latent stochastic components like trends cycles and seasonal patterns.

unpatterned wafer inspection, metrology

Inspect bare wafers.

unplanned downtime, manufacturing operations

Unplanned downtime is unexpected equipment stoppage from failures or issues.

unplanned maintenance, production

Emergency repairs.

unscented kalman, time series models

Unscented Kalman Filter propagates uncertainty through nonlinear transformations using deterministic sampling.

unscheduled downtime,production

Unexpected tool failures.

unscheduled maintenance, manufacturing operations

Unscheduled maintenance responds to unexpected equipment problems.

unstructured pruning, model optimization

Unstructured pruning removes individual weights creating sparse networks.

unstructured pruning,model optimization

Remove individual weights creating sparse tensors.

unsupervised domain adaptation,transfer learning

Adapt without target labels.

up-sampling, training

Use data more than once.

upcycle / downcycle,industry

Periods of high demand/prices vs low demand/prices.

update functions, graph neural networks

Update functions in GNNs combine aggregated neighbor information with node features to compute new representations.

upf (unified power format),upf,unified power format,design

IEEE standard for power intent.

uph (units per hour),uph,units per hour,production

Wafers or lots processed per hour productivity metric.

upper confidence bound (ucb),upper confidence bound,ucb,reinforcement learning

Optimistic exploration strategy.

upper control limit, ucl, spc

Upper boundary for normal variation.

upper specification limit, usl, spc

Maximum acceptable value.

upscale,super resolution,enhance

Super resolution upscales images. AI adds detail. Real-ESRGAN, Topaz popular tools.