Introduction Interconnect strategy, or interconnect planning has become a critical part of chip design: the growing significance of wire delay relative to gate delay . increasing power consumption of wires: could be up to 50% of the total dynamic power. Influences the ASIC design methodology Our work Try to revamp the on-chip local interconnect configuration for multi-objective optimization:
Compare different objective functions. Formulate various matrics to measure the wire performance. Identify the optimal wire configurations.
Evaluation Approach and Models objective functions:
is the wire-length normalized delay. is the wire-length-normalized power.
n delay n n delay power 2 n n delay power n delay n power Evaluation Approach and Models Metrics
is amount of data that can be transferred per unit area per unit time. n delay n power bandwidth / bandwidth power ( ) 1/ n bandwidth delay pitch = Evaluation Approach and Models Models Elmore delay model:
Power model:
Leakage factor:
0 0 (1 )(1 ) (1 ) stage nmos w n w w inv w nmos inv inv inv inv d b g f r c br c delay ar c l b g r c s l l s + + = = + + + + 2 (1.1 )(1 )(1 ) stage leak nmos inv n w dd inv inv p f g c s power c v l l q | | + + + = = + | \ . Minimum Delay Optimum repeater interval and size:
Optimum delay and power Experimental results: wire configuration 0 0.2 0.4 0.6 0.8 1 1.2 x 10 -6 0 20 40 60 80 100 120 140 min-d min-ddp min-dp pitch(m) i n v e r t e r
s i z e 180nm 130nm 100nm 70nm 0 0.2 0.4 0.6 0.8 1 1.2 x 10 -6 0 1 2 3 4 5 6 7 8 9 x 10 -7 min-d min-ddp min-dp pitch(m) w i d t h ( m ) 180nm 130nm 100nm 70nm Optimal inverter sizes Optimal widths 0 0.2 0.4 0.6 0.8 1 1.2 x 10 -6 0 0.5 1 1.5 2 2.5 x 10 -3 pitch(m) i n v e r t e r
d i s t a n c e ( m ) 180nm 130nm 100nm 70nm 0 0.2 0.4 0.6 0.8 1 1.2 x 10 -6 0 0.5 1 1.5 2 2.5 3 x 10 -3 pitch(m) i n v e r t e r
d i s t a n c e ( m ) 180nm 130nm 100nm 70nm 0 0.2 0.4 0.6 0.8 1 1.2 x 10 -6 0 0.5 1 1.5 2 2.5 3 x 10 -3 pitch(m) i n v e r t e r
d i s t a n c e ( m ) 180nm 130nm 100nm 70nm Optimal inverter distances in the min-dp procedure Optimal inverter distances in the min-ddp procedure Optimal inverter distances in the min-d procedure Experimental results: metric evaluation 50 100 150 200 0 0.5 1 1.5 x 10 -6 0 0.5 1 1.5 2 x 10 -7 te c h n o d e p it c h ( m ) d e l a y n ( s / m ) min-d min-ddp min-dp 50 100 150 200 0 1 2 x 10 -6 0 1 2 3 4 5 6 x 10 -10 te ch n o d e p i t c h ( m ) p o w e r ( J / m ) min-d min-ddp min-dp n delay n power 50 100 150 200 0 0.5 1 1.5 x 10 -6 1 2 3 4 5 x 10 13 te c h n o d e p itc h (m ) b a n d w i d t h ( b i t s / s ) min-d min-ddp min-dp 50 100 150 200 0 0.5 1 1.5 x 10 -6 0 1 2 3 4 5 x 10 23 t e c h n o d e p itc h (m ) b a n d w i d t h / p o w e r ( m / J s ) min-d min-ddp min-dp Experimental results: metric evaluation bandwidth / bandwidth power