![Page 1: 1 Design Space Exploration for Power-Efficient Mixed-Radix Ling Adders Chung-Kuan Cheng Computer Science and Engineering Depart. University of California,](https://reader035.vdocuments.mx/reader035/viewer/2022062802/56649ea45503460f94ba8929/html5/thumbnails/1.jpg)
1
Design Space Exploration for Power-Efficient Mixed-Radix Ling Adders
Chung-Kuan ChengComputer Science and Engineering
Depart. University of California, San Diego
![Page 2: 1 Design Space Exploration for Power-Efficient Mixed-Radix Ling Adders Chung-Kuan Cheng Computer Science and Engineering Depart. University of California,](https://reader035.vdocuments.mx/reader035/viewer/2022062802/56649ea45503460f94ba8929/html5/thumbnails/2.jpg)
2
Outline Prefix Adder Problem
Background & Previous Work Extensions: High-radix, Ling
Our Work Area/Timing/Power Models Mixed-Radix (2,3,4) Adders ILP Formulation
Experimental Results Future Work
![Page 3: 1 Design Space Exploration for Power-Efficient Mixed-Radix Ling Adders Chung-Kuan Cheng Computer Science and Engineering Depart. University of California,](https://reader035.vdocuments.mx/reader035/viewer/2022062802/56649ea45503460f94ba8929/html5/thumbnails/3.jpg)
Prefix Adder – Challenges
Logical Levels
Wire Tracks
Fanouts
Area
Physical placement
Detail routing
New Design Scope
Timing
Gate Cap
Wire Cap
Gate sizingBuffer insertion
Signal slope
Input arrival time
Output require time
Increasing impact of physical design and concern of power.
Power
Static power
Dynamic power
Power gating
Activity Probability
![Page 4: 1 Design Space Exploration for Power-Efficient Mixed-Radix Ling Adders Chung-Kuan Cheng Computer Science and Engineering Depart. University of California,](https://reader035.vdocuments.mx/reader035/viewer/2022062802/56649ea45503460f94ba8929/html5/thumbnails/4.jpg)
4
Binary Addition Input: two n-bit binary numbers
and , one bit carry-in Output: n-bit sum and one bit
carry out Prefix Addition: Carry generation &
propagation
011... aaan
011... bbbn 0c
011... sssn
nc
)(
:Propagate
:Generate
1
iiii
iiii
iii
iii
bacs
cpgc
bap
bag
![Page 5: 1 Design Space Exploration for Power-Efficient Mixed-Radix Ling Adders Chung-Kuan Cheng Computer Science and Engineering Depart. University of California,](https://reader035.vdocuments.mx/reader035/viewer/2022062802/56649ea45503460f94ba8929/html5/thumbnails/5.jpg)
5
Prefix Addition – Formulation
iiiiii bapbag Pre-processing:
Post-processing:
Prefix Computation:
iii
iii
cps
cPGc
0]0:[]0:[1
]:1[]:[]:[
]:1[]:[]:[]:[
kjjiki
kjjijiki
PPP
GPGG
![Page 6: 1 Design Space Exploration for Power-Efficient Mixed-Radix Ling Adders Chung-Kuan Cheng Computer Science and Engineering Depart. University of California,](https://reader035.vdocuments.mx/reader035/viewer/2022062802/56649ea45503460f94ba8929/html5/thumbnails/6.jpg)
6
1234
12:13:14:1
Prefix Adder – Prefix Structure Graph
gpi
pi
G[i:0]
si
biai
GP[i, j] GP[j-1, k]
GP[i, k]
gp generator
sum generator
GP cell
Pre-processing
Post-processing
Prefix Computation
![Page 7: 1 Design Space Exploration for Power-Efficient Mixed-Radix Ling Adders Chung-Kuan Cheng Computer Science and Engineering Depart. University of California,](https://reader035.vdocuments.mx/reader035/viewer/2022062802/56649ea45503460f94ba8929/html5/thumbnails/7.jpg)
7
Previous Works – Classical prefix adders
12345678
12:13:14:16:17:18:1 5:1
Brent-Kung:Logical levels: 2log2n–1 Max fanouts: 2Wire tracks: 1
12345678
12:13:14:16:17:18:1 5:1
Kogge-Stone:Logical levels: log2nMax fanouts: 2Wire tracks: n/2
12345678
12:13:14:16:17:18:1 5:1
Sklansky:Logical levels: log2nMax fanouts: n/2Wire tracks: 1
![Page 8: 1 Design Space Exploration for Power-Efficient Mixed-Radix Ling Adders Chung-Kuan Cheng Computer Science and Engineering Depart. University of California,](https://reader035.vdocuments.mx/reader035/viewer/2022062802/56649ea45503460f94ba8929/html5/thumbnails/8.jpg)
8
High-Radix Adders
Each cell has more than two fan-in’s
Pros: less logic levels 6 levels (radix-2) vs. 3 levels (radix-4)
for 64-bit addition Cons: larger delay and power in
each cell
![Page 9: 1 Design Space Exploration for Power-Efficient Mixed-Radix Ling Adders Chung-Kuan Cheng Computer Science and Engineering Depart. University of California,](https://reader035.vdocuments.mx/reader035/viewer/2022062802/56649ea45503460f94ba8929/html5/thumbnails/9.jpg)
9
Radix-3 Sklansky & Kogge-Stone Adder
David Harris, “Logical Effort of Higher Valency Adders”
![Page 10: 1 Design Space Exploration for Power-Efficient Mixed-Radix Ling Adders Chung-Kuan Cheng Computer Science and Engineering Depart. University of California,](https://reader035.vdocuments.mx/reader035/viewer/2022062802/56649ea45503460f94ba8929/html5/thumbnails/10.jpg)
10
Ling Adders
iiiiii bapbag Pre-processing:
Post-processing:
Prefix Computation:
iii
iii
cps
cPGc
0]0:1[]0:1[
]:1[]:[]:[
]:1[]:[]:[]:[
kjjiki
kjjijiki
PPP
GPGG
Prefix Ling
* *[ 1:0] [ 1:0] 1( )i i i i i is G t G t p
,i i i i i i
i i i
g a b p a b
t a b
1*
]1:[1*
]1:[ , iiiiiiii ppPggG* * * *[ : ] [ : ] [ 1: 1] [ 1: ]i k i j i j j kG G P G
*]:1[
*]:[
*]:[ kjjiki PPP
*]0:1[1 iii Gpc
![Page 11: 1 Design Space Exploration for Power-Efficient Mixed-Radix Ling Adders Chung-Kuan Cheng Computer Science and Engineering Depart. University of California,](https://reader035.vdocuments.mx/reader035/viewer/2022062802/56649ea45503460f94ba8929/html5/thumbnails/11.jpg)
11
An 8-bit Ling Adder
H1H2H3H4H5H6H7H8
12345678
0 1
si
ai bipi-1
Hi
gi pi-1 pi-2
G*i P*
i-1
gi-1
![Page 12: 1 Design Space Exploration for Power-Efficient Mixed-Radix Ling Adders Chung-Kuan Cheng Computer Science and Engineering Depart. University of California,](https://reader035.vdocuments.mx/reader035/viewer/2022062802/56649ea45503460f94ba8929/html5/thumbnails/12.jpg)
12
Area Model
Distinguish physical placement from logical structure, but keep the bit-slice structure.
Logical view Physical view
Bit position
Logical level
Bit position
Physical level
Compact placement
12345678 12345678
![Page 13: 1 Design Space Exploration for Power-Efficient Mixed-Radix Ling Adders Chung-Kuan Cheng Computer Science and Engineering Depart. University of California,](https://reader035.vdocuments.mx/reader035/viewer/2022062802/56649ea45503460f94ba8929/html5/thumbnails/13.jpg)
13
Timing Model Cell delay calculation:
pfd Effort Delay Intrinsic Delay
hgf
Logical EffortElectrical Effort = Cout/Cin= (fanouts+wirelength) / size
Intrinsic properties of the cell
![Page 14: 1 Design Space Exploration for Power-Efficient Mixed-Radix Ling Adders Chung-Kuan Cheng Computer Science and Engineering Depart. University of California,](https://reader035.vdocuments.mx/reader035/viewer/2022062802/56649ea45503460f94ba8929/html5/thumbnails/14.jpg)
14
Power Model Total power consumption:
Dynamic power + Static Power Static power: leakage current of device
Psta = *#cells Dynamic power: current switching
capacitancePdyn = Cload
is the switching probability = j (j is the logical level*)
cellsCjPPP loadstadyntotal # * Vanichayobon S, etc, “Power-speed Trade-off in Parallel Prefix Circuits”
![Page 15: 1 Design Space Exploration for Power-Efficient Mixed-Radix Ling Adders Chung-Kuan Cheng Computer Science and Engineering Depart. University of California,](https://reader035.vdocuments.mx/reader035/viewer/2022062802/56649ea45503460f94ba8929/html5/thumbnails/15.jpg)
15
ILP Formulation OverviewStructure variables: •GP cells•Connections (wires)•Physical positions
Capacitance variables: •Gate cap•Vertical wire cap•Horizontal wire cap
Timing variables: •Input arrival time•Output arrival time
Power Objective
ILPILOG CPLEX
Optimal Solution
![Page 16: 1 Design Space Exploration for Power-Efficient Mixed-Radix Ling Adders Chung-Kuan Cheng Computer Science and Engineering Depart. University of California,](https://reader035.vdocuments.mx/reader035/viewer/2022062802/56649ea45503460f94ba8929/html5/thumbnails/16.jpg)
16
Integer Linear Programming (ILP) ILP: Linear Programming with integer
variables. Difficulties and techniques:
Constraints are not linear Linearize using pseudo linear constraints
Search Space too large Reduce search space
Search is slow Add redundant constraints to speedup
![Page 17: 1 Design Space Exploration for Power-Efficient Mixed-Radix Ling Adders Chung-Kuan Cheng Computer Science and Engineering Depart. University of California,](https://reader035.vdocuments.mx/reader035/viewer/2022062802/56649ea45503460f94ba8929/html5/thumbnails/17.jpg)
17
ILP – Integer Linear Programming
Linear Programming: linear constraints, linear objective, fractional variables.
Integer Linear Programming: Linear Programming with integer variables.
LP Optimal
Constraints
ILP Optimal
![Page 18: 1 Design Space Exploration for Power-Efficient Mixed-Radix Ling Adders Chung-Kuan Cheng Computer Science and Engineering Depart. University of California,](https://reader035.vdocuments.mx/reader035/viewer/2022062802/56649ea45503460f94ba8929/html5/thumbnails/18.jpg)
ILP – Pseudo-Linear Constraint
Minimize: x3
Subject to: x1 300
x2 500
x3 = min(x1, x2)
Minimize: x3
Subject to: x1 300
x2 500
x3 x1
x3 x2
x3 x1 – 1000 b1 (1)
x3 x2 – 1000 (1 – b1) (2)
b1 is binary
Problem: ILP formulation:
LP objective: 0ILP objective: 300
• A constraint is called pseudo-linear if it’s not effective until some integer variables are fixed.
• Pseudo-linear constraints mostly arise from IF/ELSE scenarios• binary decision variables are introduced to indicate true or false.
![Page 19: 1 Design Space Exploration for Power-Efficient Mixed-Radix Ling Adders Chung-Kuan Cheng Computer Science and Engineering Depart. University of California,](https://reader035.vdocuments.mx/reader035/viewer/2022062802/56649ea45503460f94ba8929/html5/thumbnails/19.jpg)
19
ILP Solver Search Procedure
0Root (all vars are fractional)
b1
b2
b3
b4
2
‘0’ ‘1’infeasible
‘0’ ‘1’
3feasible
(current best)
infeasible
Minimize F(b1, b2, b3, b4, f1,…) bi is binary
It is VERY helpful if ILP objective is close to LP objective
2
‘0’ ‘1’
4
‘0’ ‘1’
3 2
4
55
‘0’ ‘1’
Cut
2 Bound (Smallest candidate)
![Page 20: 1 Design Space Exploration for Power-Efficient Mixed-Radix Ling Adders Chung-Kuan Cheng Computer Science and Engineering Depart. University of California,](https://reader035.vdocuments.mx/reader035/viewer/2022062802/56649ea45503460f94ba8929/html5/thumbnails/20.jpg)
Interval Adjacency Constraint
H1H2H3H4H5H6H7H8
12345678
(7,3): Interval [7,1]
(3,2): Interval [3,1]
(7,2): Interval [7,4]
Must be adjacent,i.e. 4 = 3 + 1
(column id, logic level)
![Page 21: 1 Design Space Exploration for Power-Efficient Mixed-Radix Ling Adders Chung-Kuan Cheng Computer Science and Engineering Depart. University of California,](https://reader035.vdocuments.mx/reader035/viewer/2022062802/56649ea45503460f94ba8929/html5/thumbnails/21.jpg)
21
Linearization for Interval Adjacency Constraint
(i, j)
(i, h) (k1, l1) (k2, l2)
wl wr1 wr2
],[ ),(),(R
hiL
hi yy ],[ )1,1()1,1(R
lkL
lk yy ],[ )2,2()2,2(R
lkL
lk yy
],[ ),(),(R
jiL
ji yy
11 if 1),(),( (i,j,k,l) wrwl(i,j,h) yy Llk
Rhi
1 if 1),,,(1),(
),( wl(i,j,h) lkjiwrkylk
Rhi
11 ),,,(1),(
),( wl(i,j,h))(nlkjiwrkylk
Rhi
11 ),,,(1),(
),( wl(i,j,h))(nlkjiwrkylk
Rhi
iyLji ),(
Linearize
Pseudo Linear
Left interval bound equal to column index
![Page 22: 1 Design Space Exploration for Power-Efficient Mixed-Radix Ling Adders Chung-Kuan Cheng Computer Science and Engineering Depart. University of California,](https://reader035.vdocuments.mx/reader035/viewer/2022062802/56649ea45503460f94ba8929/html5/thumbnails/22.jpg)
22
Search Space Reduction Ling’s adder:
separate odd and even bits
Double the bit-width we are able to search H1H2H3H4H5H6H7H8
12345678
![Page 23: 1 Design Space Exploration for Power-Efficient Mixed-Radix Ling Adders Chung-Kuan Cheng Computer Science and Engineering Depart. University of California,](https://reader035.vdocuments.mx/reader035/viewer/2022062802/56649ea45503460f94ba8929/html5/thumbnails/23.jpg)
23
Redundant Constraints Cell (i,j) is known to have logic level j before
wire connection Assume load is MinLoad (fanout=1 with
minimum wire length):
MinLoadjP ji ),(
Cell (i,j) has a path of length j-1 Assume each cell along the path has MinLoad
)(),( MinLoadLEPDjT ji
![Page 24: 1 Design Space Exploration for Power-Efficient Mixed-Radix Ling Adders Chung-Kuan Cheng Computer Science and Engineering Depart. University of California,](https://reader035.vdocuments.mx/reader035/viewer/2022062802/56649ea45503460f94ba8929/html5/thumbnails/24.jpg)
24
Experiments – 16-bit Uniform Timing
![Page 25: 1 Design Space Exploration for Power-Efficient Mixed-Radix Ling Adders Chung-Kuan Cheng Computer Science and Engineering Depart. University of California,](https://reader035.vdocuments.mx/reader035/viewer/2022062802/56649ea45503460f94ba8929/html5/thumbnails/25.jpg)
Experiments – 16-bit Uniform Timing
25
![Page 26: 1 Design Space Exploration for Power-Efficient Mixed-Radix Ling Adders Chung-Kuan Cheng Computer Science and Engineering Depart. University of California,](https://reader035.vdocuments.mx/reader035/viewer/2022062802/56649ea45503460f94ba8929/html5/thumbnails/26.jpg)
26
Min-Power Radix-2 Adder (delay= 22, power = 45.5FO4 )
1
1
2
2
3
3
4
4
5
5
6
6
7
7
8
8
9
9
10
10
11
11
12
12
13
13
14
14
15
15
16
16
![Page 27: 1 Design Space Exploration for Power-Efficient Mixed-Radix Ling Adders Chung-Kuan Cheng Computer Science and Engineering Depart. University of California,](https://reader035.vdocuments.mx/reader035/viewer/2022062802/56649ea45503460f94ba8929/html5/thumbnails/27.jpg)
27
Min-Power Radix-2&4 Adder (delay=18, power = 29.75FO4 )
1
1
2
2
3
3
4
4
5
5
6
6
7
7
8
8
9
9
10
10
11
11
12
12
13
13
14
14
15
15
16
16
Radix-2 Cell Radix-4 Cell
![Page 28: 1 Design Space Exploration for Power-Efficient Mixed-Radix Ling Adders Chung-Kuan Cheng Computer Science and Engineering Depart. University of California,](https://reader035.vdocuments.mx/reader035/viewer/2022062802/56649ea45503460f94ba8929/html5/thumbnails/28.jpg)
28
Min-Power Mixed-Radix Adder (delay=20, power = 28.0FO4)
1
1
2
2
3
3
4
4
5
5
6
6
7
7
8
8
9
9
10
10
11
11
12
12
13
13
14
14
15
15
16
16
Radix-2 Cell Radix-4 Cell Radix-3 Cell
![Page 29: 1 Design Space Exploration for Power-Efficient Mixed-Radix Ling Adders Chung-Kuan Cheng Computer Science and Engineering Depart. University of California,](https://reader035.vdocuments.mx/reader035/viewer/2022062802/56649ea45503460f94ba8929/html5/thumbnails/29.jpg)
29
Experiments – 16-bit Non-uniform Time (Mixed Radix)
ILP is able to handle non-uniform timings
Ling adders are most superior in increasing arrival time– faster carries
![Page 30: 1 Design Space Exploration for Power-Efficient Mixed-Radix Ling Adders Chung-Kuan Cheng Computer Science and Engineering Depart. University of California,](https://reader035.vdocuments.mx/reader035/viewer/2022062802/56649ea45503460f94ba8929/html5/thumbnails/30.jpg)
Increasing Arrival Time (delay=35.5, power = 27.0FO4 )
30
![Page 31: 1 Design Space Exploration for Power-Efficient Mixed-Radix Ling Adders Chung-Kuan Cheng Computer Science and Engineering Depart. University of California,](https://reader035.vdocuments.mx/reader035/viewer/2022062802/56649ea45503460f94ba8929/html5/thumbnails/31.jpg)
Decreasing Arrival Time (delay=34.5, power = 30.5FO4)
31
![Page 32: 1 Design Space Exploration for Power-Efficient Mixed-Radix Ling Adders Chung-Kuan Cheng Computer Science and Engineering Depart. University of California,](https://reader035.vdocuments.mx/reader035/viewer/2022062802/56649ea45503460f94ba8929/html5/thumbnails/32.jpg)
Convex Arrival Time (delay=35.9, power = 32.4FO4 )
32
![Page 33: 1 Design Space Exploration for Power-Efficient Mixed-Radix Ling Adders Chung-Kuan Cheng Computer Science and Engineering Depart. University of California,](https://reader035.vdocuments.mx/reader035/viewer/2022062802/56649ea45503460f94ba8929/html5/thumbnails/33.jpg)
Increasing Required Time (delay=34.5, power = 30.5FO4)
33
![Page 34: 1 Design Space Exploration for Power-Efficient Mixed-Radix Ling Adders Chung-Kuan Cheng Computer Science and Engineering Depart. University of California,](https://reader035.vdocuments.mx/reader035/viewer/2022062802/56649ea45503460f94ba8929/html5/thumbnails/34.jpg)
Decreasing Required Time (delay=36.5, power = 32.5FO4)
34
![Page 35: 1 Design Space Exploration for Power-Efficient Mixed-Radix Ling Adders Chung-Kuan Cheng Computer Science and Engineering Depart. University of California,](https://reader035.vdocuments.mx/reader035/viewer/2022062802/56649ea45503460f94ba8929/html5/thumbnails/35.jpg)
Convex Required Time (delay=36.5, power = 32.5FO4)
35
![Page 36: 1 Design Space Exploration for Power-Efficient Mixed-Radix Ling Adders Chung-Kuan Cheng Computer Science and Engineering Depart. University of California,](https://reader035.vdocuments.mx/reader035/viewer/2022062802/56649ea45503460f94ba8929/html5/thumbnails/36.jpg)
36
Experiments – 64-bit Hierarchical Structure (Mixed-Radix)
Handle high bit-width applications 16x4 and 8x8
ILP Block ILP Block ILP Block ILP Block
ILP Block
a1b1a16b16a17b17a32b32a33b33a48b48a49b49a64b64
…... …... …... …...
Level 1
Level 2
…... …... …... …...
…... …... …... …...
GP*[64:50]GP*[48:34] GP*[32:18] GP*[16:2]
GP*[1:1]GP*[17:17]GP*[33:33]GP*[49:49]
…... …... …...H64 H49 H48 H33 H32 H17 H16 H1
![Page 37: 1 Design Space Exploration for Power-Efficient Mixed-Radix Ling Adders Chung-Kuan Cheng Computer Science and Engineering Depart. University of California,](https://reader035.vdocuments.mx/reader035/viewer/2022062802/56649ea45503460f94ba8929/html5/thumbnails/37.jpg)
37
Experiments – 64-bit Hierarchical Structure
TSL: a 64-bit high-radix three-stage Ling adder
V. Oklobdzija and B. Zeydel, “Energy-Delay Characteristics of CMOS Adders”,in High-Performance Energy-Efficient Microprocessor Design, pp. 147-170, 2006
![Page 38: 1 Design Space Exploration for Power-Efficient Mixed-Radix Ling Adders Chung-Kuan Cheng Computer Science and Engineering Depart. University of California,](https://reader035.vdocuments.mx/reader035/viewer/2022062802/56649ea45503460f94ba8929/html5/thumbnails/38.jpg)
38
ASIC Implementation - Results 64-bit hierarchical design (mixed-radix)
by ILP vs. fast carry look-ahead adder by Synopsys Design Compiler
TSMC 90nm standard cell library was used
![Page 39: 1 Design Space Exploration for Power-Efficient Mixed-Radix Ling Adders Chung-Kuan Cheng Computer Science and Engineering Depart. University of California,](https://reader035.vdocuments.mx/reader035/viewer/2022062802/56649ea45503460f94ba8929/html5/thumbnails/39.jpg)
39
Future Work
ILP formulation improvement Expected to handle 32 or 64 bit
applications without hierarchical scheme
Optimizing other computer arithmetic modules Comparator, Multiplier
![Page 40: 1 Design Space Exploration for Power-Efficient Mixed-Radix Ling Adders Chung-Kuan Cheng Computer Science and Engineering Depart. University of California,](https://reader035.vdocuments.mx/reader035/viewer/2022062802/56649ea45503460f94ba8929/html5/thumbnails/40.jpg)
40
Q & A
Thank You!