Ikibazo gihuza ibibazo · Biragoye Urwego

AI muri Semiconductor Fab Yitanga Optimisation Ikibazo

Igeragezwa rya wafer ikarita yerekana ibyiciro, ibipimo bya metero bivuye mubikoresho bya sensor sensor, hamwe na run-to-run process control in chip fab umusaruro.

Ikibazo 1 Bya 8

'Umusaruro' bisobanura iki mu gukora semiconductor?

Reba buri kibazo8 ibibazo

Ibisubizo nibisobanuro bigaragara mugihe ufata ikibazo.

  1. 'Umusaruro' bisobanura iki mu gukora semiconductor?
    • The number of wafers a fab processes per day
    • The fraction of dies on a wafer that work
    • The thickness of deposited films
    • The speed of a finished chip
  2. Metrology isanzwe ikora iki?
    • Designs chip layouts automatically
    • Predicts measurement results from tool sensor data for every wafer
    • Replaces all inspection tools with cameras
    • Simulates customer demand for chips
  3. Igishushanyo-mpeta ku ikarita ya wafer birashoboka cyane kwerekana iki?
    • A problem affecting the wafer edge during a process like etching
    • A software bug in the test program
    • A random particle in the center
    • A customer design error
  4. Niki Dataset ya WM-811K ikoreshwa?
    • Training chip layout generators
    • Research on classifying wafer map defect patterns
    • Forecasting semiconductor prices
    • Measuring transistor speed
  5. Ni ukubera iki moderi yubumenyi bwa metero ikenera kwisubiramo kenshi?
    • Sensors stop recording data after a year
    • Maintenance, chamber aging and tool differences shift the data distribution
    • Regulators require monthly retraining
    • Models forget their weights overnight
  6. What does a run-to-run controller do?
    • Schedules staff shifts
    • Adjusts recipe settings between lots to keep results on target
    • Classifies SEM images of defects
    • Tests finished chips for speed
  7. Why do teams add an explicit 'unknown' class to wafer map classifiers?
    • To speed up training
    • So new patterns are flagged instead of forced into a known label
    • Kubera ko amategeko abisaba
    • To reduce the number of test wafers
  8. Which statement about fab training data is accurate?
    • Every defect type has millions of labeled examples
    • Rare defects may appear only a few times and classes are heavily imbalanced
    • Labels are generated automatically with no expert input
    • Fabs share all data publicly