QLF08_LaBel.pdf - 第30页
IC and Component Selection for Space Sy stems – Presented by Kenneth A . LaBel 30 Programmatics: Development Cost • Overall product development costs are increasing: design, mask, and manufacturing – However , the percen…
IC and Component Selection for Space Systems – Presented by Kenneth A. LaBel
29
Programmatics:
What if the first design is incorrect?
• As digital devices pack more bits into the same physical
space (i.e., technology feature size is shrinking), more
designs require a “second pass” or re-spin.
• Even worse, the time it takes for this re-spin has increased
0
1
2
3
4
5
6
7
Months
.25u .18u .13u .09u
Technology
Average Re-Spin time
Source: www.chipdesignmag.com
0
10
20
30
40
50
60
.25u .18u .13u .09u
Technology
Percent of designs requiring re-spin
IC and Component Selection for Space Systems – Presented by Kenneth A. LaBel
30
Programmatics: Development Cost
• Overall product development costs are increasing: design, mask, and
manufacturing
– However, the percentage of product cost for both design and masks (>$1M for
90nm!) are also Increasing
• Implies cost effectiveness of minimizing masks and design re-spin costs (advantage
FPGAs and SAs)
Graphic from MIPS Technologies "The Coming Reality for SOC Designers" by John Bourgoin
Design
Costs
Mask
Costs
Mfg.
Costs
$0M
$1M
$2M
$3M
$4M
$5M
$6M
$7M
$8M
$9M
$10M
1995
(0.35um)
1997
(0.25um)
1999
(0.18um)
2001
(0.13um)
2003
(0.10um)
Assumptions: 6M-70M transistors, 3 mask spins, 250K Units, transistor costs and productivity projected from the ITRS
Design
100%
Mfg.
(250KU)
62%
Masks
Design
100%
Masks
Mfg.
(250KU)
1995
13%
IC and Component Selection for Space Systems – Presented by Kenneth A. LaBel
31
Selection Criteria: Radiation and Reliability
• While there is no “generic” answer for radiation
tolerance and reliability levels in the trade, there
are numerous considerations such as
– What is known of the process radiation tolerance?
• Was process/device radiation qualification sufficient for
planned design or application?
– Library? Cells? Speed? Etc…
• Is the process reliable and achieving good yield?
– High-volume commercial fab vs. low-volume niche fab?
• Are there lot-specific or application-specific concerns?
– Is there a specific performance requirement such as
reconfigurability or ultra-low power that forces a non-
radiation tolerant device selection?
• How does a fault-tolerant system architecture enter the
equation?
– How testable is the design/device?
• As has become more evident, trade spaces are
much more complex than “just” an IC