1. A Procedure and Program to Optimize Shuttle Mask Cost Advantage
Artur Balasinski 1) , J.Cetin1), W.Sachs-Baker1)
A.Kahng2), and X.Xu2)
1) Cypress Semiconductor
2) University of California
San Diego, CA, USA

2. Introduction
Shuttle or Multi-IP masks: A well-known recipe for reducing cost of mask component in new product development.
Simple concept, complex execution:
put multiple types of layers or products on one mask plate
process options have to be considered.
Immediate verification required:
Once the first mask is shipped, it is costly to change architecture of the set.
Visual verification fast and efficient
Simple algorithm:
mask layers lined up according to the price and field tone
subsequent layers placed on the masks with different grades.
Verification and enhancements:
layer pairing based not only on the price of the mask set,
but also on the cost of the setup, design, and fab-friendliness.
Manual pairing
reduce the count of orphan layers
matching rules to improve flexibility and fab-friendly architecture.
11/23/2018

3. Why Are We Optimizing ? IP distribution:
mask databases by layer and product
can have different shapes, sizes, mask grade requirements
Optimization needed to put as many non-redundant IP components on one plate as possible
Two basic flavors:
Multi-layer (NL-1P)
Multi-Product (1L-MP)
Mixtures also possible
Gets complicated for 30 masks per product
Mask 1
High end
Starting layer
set
Mask 2
Low end
Starting
product set
11/23/2018

4. What Are We Optimizing ? Not mask cost because mask cost [$] = const
IP placement cost:
what is the min cost of reticles to transfer layout pattern to wafer,
for one product
for a product family
multiple IP needed at the same time/plate to share placement cost
IP density on a mask:
Examples:
Simple:
1 Product with N layers (1P-NL) on one mask: IPD = N (no orphans !)
M Products with 1 layer (1L-MP) on one mask: IPD = M (same flow !)
Real life:
M products with total of n1+n2+…nm layers on K masks:
IPD = (n1+n2+…nm)/ K
Why real life is complicated: IP placement for testchips costs more than for products
11/23/2018

5. What Are We Optimizing ?
The higher the IPD, the lower the mask component cost, but
Mask cost is not the only one in the overall product development cost
11/23/2018

6. cost may increase with IPD
How Are We Optimizing ?
Optimizing cost of IP placement:
Increase IPD until something breaks
Multi-component cost criteria:
First derivative:
Manufacturing
Setup
Design
cost may increase with IPD
Placement cost
decreases with IPD
Question:
Can cost reduction only depend
on automated layer placement
11/23/2018

7. What can increase with IPD ?
Setup, Design, and Manufacturing cost
single-event
distributed
As IPD increases:
Manufacturing cost up
predictable - low volume as mitigation
unpredictable - fab conditions
Setup cost often not included
database handling
frame planning
quality procedures
archiving
Design cost:
shuttle delays volume production
wasted effort for unproven products
What this means:
Full visibility on shuttle impact desired upfront
11/23/2018

8. Example Low IPD Mask cost dominates
Other components catch up after multiple weeks
High IPD
Design cost dominates
Mask cost distributed
Other cost increased
due to higher complexity
11/23/2018

9. Different ramp rate: use high IPD masks for short time only
Cumulative Cost
Low IPD
Total cost similar
for low and high IPD
Different ramp rate: use high IPD masks for short time only
High IPD
11/23/2018

10. Basic Algorithm for NL-1P
How do we arrange layers on a mask plate ?
Step 1:
Decision making
Do we want this product to run in mulit-layer format ?
With how many layers per plate ?
Output: N = number of layers per plate
11/23/2018

11. Basic Algorithm for NL-1P
Step 2:
Execution
Guidelines:
- Design feedback
- Product volume
11/23/2018

12. Input GUI Input parameters: - Mask making restrictions
- Scribe entries
- Number of LPP
- Reticle data (to be loaded)
Hint:
For best results, have a picture !
11/23/2018

13. Reticle Grade A1, Cost 29,000, Process 193-PSM-ARP
Output GUI 1
Simple example for 3 products with identical die sizes
Picture:
Layer name, blading, scribe to scale
Price incentive displayed
1 Picture = Words
Reticle Grade A1, Cost 29,000, Process 193-PSM-ARP
11/23/2018

14. Output GUI 2 Mfg verification:
1. Number of exposures must not reduce stepper’s throughput
2. Other issues related to too frequent stepping must be avoided
11/23/2018

15. Manual Optimization Layer placement procedure:
good for majority of the layers
may not be optimal from the MFG, SET or DES standpoint
enhancements, limitations and additional rules for the layer placement:
layers with similar functions should be placed on one reticle, to enable single mask retapeout for design fixes,
non-critical layers OK with the critical ones even if their characteristics, e.g., field tone, are different. The exact guidelines provided by the maskshop,
layers with similar pattern densities should be placed on the same reticle,
individual layers from different products need large footprint so as not to reduce fab throughput.
Should provide data placement document.
These rules may not lend themselves to easy automation. While the basic layer combinations call for price-dependent layer pairing, this may not always be the best solution for product development.
11/23/2018

16. Unlimited placement required Any combination of masks and layers
Theoretical Case
Unlimited placement required
Any combination of masks and layers
11/23/2018

17. Conclusions
Shuttle automation is making significant progress to optimize placement of multiple products on the masks. One consequence is complex data structure which impacts design and manufacturability.
Layer placement needs automated rules and manual verification to minimize the number of orphans and make the set manufacturing-friendly at the same time.
Algorithmic solutions aimed only at reduction of the mask cost may not provide optimal mask composition, due to many factors in the multi-parameter, integrated cost equation: setup, product line, risk of error and lost opportunity.
We developed and demonstrated the automated layer placement routine which can be useful in optimization of new product development.
11/23/2018