The Die Products User Club is co-ordinated and moderated by : Knowledge Based Technical Consultancy Ltd.
Email: dpuc@dpuc.net Tel: +44 (0) 1548 531414 Mob: + 44 (0) 7791 595217 Fax: +44 (0) 1548 531986
DPUC RSS Newsfeed
I Knew that...
No. 4
Using Spatial Analysis of Wafer Maps to Find Unreliable Die
A "Good Die in a Bad Neighbourhood" is a term used at production wafer test to describe die which have a good or passing Bin code but are located in an area of the wafer where most die have failing Bin codes. Identifying good die in bad neighbourhoods and eliminating them is used by chip suppliers to improve die quality and reliability. Where wafers have areas of good and poor processing, the technique also allows these wafers to continue in the production process instead of being scrapped as the die in the poorly processed regions will be eliminated.
Bin Wafer Map
Die with good bin codes are green and with bad bin codes are red.
The neighbourhood surrounding the gold die is outlined in blue and the number of failing die is counted.
Very simply, deciding if a good die is in a bad neighbourhood involves counting the number of failing die in the predefined area surrounding the good die and comparing the failing die count against a limit. More complex detection algorithms differentiate between die adjacent to the good die and distant die by weighting the proximity of failing die within the neighbourhood.
Neighbourhood layouts should be selected using statistical methods based on the die size, expected wafer yield and accepted probability of falsely detecting a bad neighbourhood.
If the failing die have a random spatial distribution but the die yield is low, there could be false detection of bad neighbourhoods unless the number of die contained in the neighbourhood is large. Also, if the die size is large, only adjacent die may be considered hence the number of die in the neighbourhood is small and the probability of false neighbourhood detection is high.
Bad neighbourhoods are present when the die yield is not constant across the wafer and indicate what yield engineers call "systematic yield loss". Randomly failing die should, statistically, not produce bad neighbourhoods. Systematic yield losses are usually attributable to processing problems caused by mal-functioning tools which do not process the entire wafer consistently. When a good die is found in a bad neighbourhood the process or process tool is on the edge between producing good and bad die.
Yield engineers and spatial signature analysis software packages can often identify the process step causing yield loss by the spatial signature of the failing die. These software packages may list all die involved in the signature, even when they are good die.
GDBN does not attempt to find a spatial signature to match the pattern of the failing die on a wafer, but instead decides if a specific neighbourhood of die is good or bad.
There will be an overlap between the unreliable die found using Part Average Testing and GDBN methods. PAT is typically done on the measurement results from a few tests without considering spatial information whereas GDBN considers the spatial information and the Bin result which is a coarse summary of the overall die performance.
Whether the die is sold in wafer form or as diced bare die or assembled die, the Bin wafer map needs to be altered immediately after probe to downgrade good die and subsequently used to eliminate the downgraded die from the production process.
Iain Gardner
Waferdata Limited
Waferdata Limited
Previous article
