Skip to content

Data Preparation

This chapter will guide you through the preparation of wafer data. You will get an overview of the data format used in wfmap, learn the wafer mapping definition in detail and how to merge it with other data source.

Definition

Wafer mapping definition used in this package is showed in follwing table, which might be slightly different as the terminology in your industry.

Parameter Definition Data Type
MAP_ROW y: vertical coordinate of the die in the wafermap Integer
MAP_COL x: horizontal coordinate of the die in the wafermap Integer
FF_ROW yn: vertical coordinate of flash field Integer
FF_COL xn: horizontal coordinate of flash field Integer
WIF_ROW vertical coordinate of the die in the corresponding flash field Integer
WIF_COL horizontal coordinate of the die in the corresponding flash field Integer
FF_SHAPE (max(FF_ROW),max(FF_COL)): Total Flash Fields by FF Row&Col Tuple
FF_SIZE (max(WIF_ROW),max(WIF_COL)): Total Dies in a Flash Field by Die Row&Col Tuple

FF is short for FlashField, WIF stands for Within FlashField.

Warning

First two parameters are required in wafer heatmap related plots and succedent 4 parameters are need for trends charts related plots. So please make sure that you have either transformed your data in similar format or pass the correct parameters to corresponding functions.

Below is an illustration for above definition.

Wafer Mapping Definition

Sample Data

Sample data is shipped within this packages, you can load it as following snippet:

from wfmap.data import load_data
data=load_data()

Besides the first 6 columns which define the wafer mapping information, additional 1 categorical feature DEFECT & 4 numerial features(MR,HR,HDI,MR2) are included for demostration. 

data.head()

Sample Data

Merge Other Data

In most common scenario, wafer data from other source is processed and stored separately by unique DeviceID such as OCR, rather than wafer data MAP_ROW & MAP_COL. Since wafer map definition is not included, You’ll need to merge it with wafer map data manually. In my practice, I saved the wafer mapping data as pickle file by different wafer type in advance. Then can load it from local directory, no need to read corresponding wafer data from file server or database.

Here’s a sample snippets for your reference, which enable to merge the wafer map data from either local pickle file or extract it from wafer incoming data.

from wfdata import Wafer # Load for Wafer Incoming Data
def merge_wfmap(df, ocr_col='SLIDER_OCR_NO', mode='AUTO', join='outer', inc_data=False, map_dir='D:\MAP_DATA'):
    """
    `SLIDER_OCR_NO` = `Wafer` + `OCR`
    """
    assert mode in ['AUTO', 'UP', 'UP2', 'UP3', 'UP2S',
                    'UP3S'], print('Wafer Format Not Suppported!!')
    assert join in ['inner', 'outer'], print('Merge Method Not Suppported!!')

    df['Wafer'] = df[ocr_col].map(lambda x: x[:5])
    df['OCR'] = df[ocr_col].map(lambda x: x[-5:])
    df.set_index('OCR', inplace=True)
    result = []
    for wf, dff in df.groupby('Wafer'):
        if mode == 'AUTO':
            if inc_data:
                map_data = Wafer("H"+wf).inc_data()
            else:
                map_data = Wafer("H"+wf).map_data()
        else:
            map_data = pd.read_pickle(f'{map_dir}/{mode}')
        merged = pd.merge(map_data, dff, right_index=True,
                          left_index=True, how=join)
        merged = merged.drop_duplicates(subset=['MAP_ROW', 'MAP_COL'])
        result.append(merged)
    return pd.concat(result)