.streamlit/secrets.toml

HF_API_KEY = '[... HuggingFace API key if running HuggingFace-hosted models ...]'

OPENAI_API_KEY = '[... OpenAI API key if running on OpenAI services ...]'

streamlit-run:

@poetry run streamlit run streamlit-main.py --server.allowRunOnSave=true --server.runOnSave=true

Install OpenSSA repo beforehand by running `make install` (or `.\make install` on Windows)

at the OpenSSA repo root directory.

In this example's directory:

- Necessary credentials need to be in the `.env` file

- DANA problem-solving can be run by `make dana-solve prob="..."` (or `.\make dana-solve "..."` on Windows)

- Streamlit app can be run by `make streamlit-run` (or `.\make streamlit-run` on Windows)

from argparse import ArgumentParser

from functools import cache

from pathlib import Path

from dotenv import load_dotenv

import yaml

from openssa import DANA, ProgramStore, HTP, HTPlanner, FileResource, HuggingFaceLM

from semikong_lm import SemiKongLM

load_dotenv()

BASE_DIR: Path = Path(__file__).parent

DATA_DIR_PATH: Path = BASE_DIR / 'data'

EXPERTISE_DIR_PATH: Path = BASE_DIR / 'expertise'

EXPERT_KNOWLEDGE_FILE_PATH: Path = EXPERTISE_DIR_PATH / 'expert-knowledge.txt'

with open(file=EXPERT_KNOWLEDGE_FILE_PATH,

buffering=-1,

encoding='utf-8',

errors='strict',

newline=None,

closefd=True,

opener=None) as f:

EXPERT_KNOWLEDGE: str = f.read()

EXPERT_PROGRAMS_FILE_PATH: Path = EXPERTISE_DIR_PATH / 'expert-programs.yml'

with open(file=EXPERT_PROGRAMS_FILE_PATH,

buffering=-1,

encoding='utf-8',

errors='strict',

newline=None,

closefd=True,

opener=None) as f:

EXPERT_PROGRAMS: dict[str, dict] = yaml.safe_load(stream=f)

def get_or_create_dana(use_semikong_lm: bool = True, max_depth=2, max_subtasks_per_decomp=4) -> DANA:

lm = (SemiKongLM if use_semikong_lm else HuggingFaceLM).from_defaults()

program_store = ProgramStore(lm=lm)

if EXPERT_PROGRAMS:

for program_name, htp_dict in EXPERT_PROGRAMS.items():

htp = HTP.from_dict(htp_dict)

program_store.add_or_update_program(name=program_name, description=htp.task.ask, program=htp)

return DANA(knowledge={EXPERT_KNOWLEDGE},

program_store=program_store,

programmer=HTPlanner(lm=lm, max_depth=max_depth, max_subtasks_per_decomp=max_subtasks_per_decomp),

resources={FileResource(path=DATA_DIR_PATH, re_index=True)})

if __name__ == '__main__':

arg_parser = ArgumentParser()

arg_parser.add_argument('problem')

args = arg_parser.parse_args()

print(get_or_create_dana().solve(problem=args.problem))

WaterBatch: 1, 2, 3, 4, 5, 6

ReflectedPower_W: 36.6, 35.4, 37.5, 40.1, 42.3, 37.9

ForwardPower_W: 60.0, 58.0, 58.2, 57.8, 65.3, 60.0

ChamberPressure_Torr: 3.89, 4.18, 3.85, 4.22, 4.25, 3.53

GasFlowRate_sccm: 59.7, 58.1, 56.2, 55.1, 59.7, 54.4

ChamberWallTemperature_C: 76.3, 78.3, 76.6, 79.6, 81.7, 81.9

SEMICONDUCTOR ETCHING EQUIPMENT MONITORING, MAINTENANCE & OPTIMIZATION

======================================================================

RF Power Fluctuation Rule

-------------------------

HEURISTIC: If the reflected power in the TEL Tactras RLSA Etcher increases by more than 5% for three consecutive wafers,

and the forward power adjustment exceeds 10%, this suggests possible electrode erosion or plasma instability.

ACTION: Alert the maintenance engineer to inspect the electrodes and adjust gas flow or power settings,

potentially avoiding costly equipment downtime.

Pressure Stability Monitoring

-----------------------------

HEURISTIC: If the chamber pressure deviation exceeds ±3% for more than 30 seconds during steady-state etching,

coupled with a gas flow variation of more than 5%, it indicates potential chamber wall coating or residue buildup.

ACTION: Recommend a preventative chamber clean and adjust the gas flow to stabilize pressure,

reducing the likelihood of non-uniform etching.

Temperature-Dependent Chamber Cleaning Optimization

---------------------------------------------------

HEURISTIC: If the chamber wall temperature shows a steady increase of 2°C over three lots,

despite maintaining the same power settings, it indicates polymer build-up on the chamber walls.

ACTION: Suggest adjusting the chamber clean cycle frequency and alerts engineers to check the wall lining.

Automated cleaning recommendations can significantly reduce variation and increase yield.

Etch Rate Uniformity Rule

-------------------------

HEURISTIC: If the etch rate between the wafer center and edge varies by more than 7% for three consecutive lots,

this could indicate gas flow distribution issues or hardware alignment issues within the TEL Tactras system.

ACTION: Alert the process engineer to investigate gas flow uniformity

and suggests optimizing flow parameters or performing a hardware realignment.

Chamber Matching Network Heuristic

----------------------------------

HEURISTIC: If the chamber matching network requires more than two adjustments per wafer run

to maintain desired RF power settings, it suggests possible impedance mismatch or degradation in the matching components.

ACTION: Optimize matching network settings and recommend maintenance

if impedance mismatch trends persist, potentially preventing plasma stability issues.

Microloading Effect Detection

-----------------------------

HEURISTIC: If the etch rate is higher in areas with smaller feature densities compared to high-density areas

(indicative of the microloading effect), and the difference exceeds a set threshold,

this can lead to non-uniform etching and yield loss.

ACTION: Alert engineers to optimize gas flow and power settings based on specific wafer feature density,

enhancing etch uniformity across wafers.

Real-Time Recipe Adjustment Based on Yield Data

-----------------------------------------------

HEURISTIC: If yield data shows a consistent defect pattern (e.g., at the wafer edge or specific die locations)

for more than 5% of wafers, correlate this with historical process parameters and makes real-time recipe adjustments.

ACTION: Automated recipe optimization based on yield trends can reduce defect occurrence,

improving overall productivity and wafer quality.

Anomaly Detection with Historical Baseline Comparison

-----------------------------------------------------

HEURISTIC: If real-time sensor data (e.g., temperature, pressure, RF power) shows deviations

outside historical baseline ranges for similar recipes, this could indicate tool degradation or unexpected chamber behavior.

ACTION: Trigger alerts and suggest corrective actions such as parameter adjustment or additional inspections,

preventing potential faults.

Particle Generation Control for Bevel Etching

---------------------------------------------

HEURISTIC: If particle count in the bevel area increases beyond acceptable limits during bevel etching

in the Tactras UDEMAE system, this suggests non-uniform edge plasma distribution or excessive material removal.

ACTION: Recommend reducing RF power or modifying gas flow in the bevel area,

significantly reducing the risk of yield-impacting particle contamination.

Multi-Parameter Fault Isolation for Complex Equipment

-----------------------------------------------------

HEURISTIC: If multiple parameters (e.g., RF power, pressure, temperature) show simultaneous deviations

outside standard operating windows, use historical fault isolation data to pinpoint the most likely root cause.

ACTION: Provide a ranked list of potential root causes with confidence levels,

enabling faster and more accurate fault isolation.

rf-power-fluctuation-monitoring:

task: |-

sub-htps:

- task: >-

- task: |-

- task: >-

- task: |-

- task: |-

pressure-instability-monitoring:

task: |-

sub-htps:

- task: >-

- task: |-

- task: >-

- task: |-

- task: |-

chamber-temperature-monitoring:

task: |-

sub-htps:

- task: >-

- task: |-

- task: |-