Transcript FFT
Graduate Institute of Electronics Engineering, NTU
102-1 Under-Graduate Project
Improving Timing, Area, and Power
Speaker: 黃乃珊
Adviser: Prof. An-Yeu Wu
Date: 2013/12/12
ACCESS IC LAB
ACCESS IC LAB
Graduate Institute of Electronics Engineering, NTU
Introduction
When design in RTL, the designer need to be
aware of timing, area and power issues.
Meeting timing is the most critical goal in
design. Only optimize for power or area after
timing is met.
Synthesis tools operate in gate level, and
cannot resolve all timing, area and power
issues.
P. 2
ACCESS IC LAB
Graduate Institute of Electronics Engineering, NTU
Timing Issues
Timing v.s. Performance
Latency
How long does it take to complete a particular operation?
Unit : ns, us, ms
Throughput
How many operations can be completed per second?
Unit : Mbps, Gbps
Input
1bit output
Latency = 2 clock cycles
Throughput = 1bit/clock cycle
P. 3
ACCESS IC LAB
Graduate Institute of Electronics Engineering, NTU
Timing Requirement
To fit system throughput, the timing (clock
period) must be smaller than some value.
In IC design industry, the design must meet
timing with margin, and using worst-case
library model.
P. 4
ACCESS IC LAB
Graduate Institute of Electronics Engineering, NTU
How to Improve Timing?
While your clock cycle time does not fit system
specification
Pipelining : Exploits temporal parallelism
Reduce the clock cycle time (clock period)
Insert pipeline registers without changing coherence of the
data.
Parallel skill
Decrease data flow of your design
P. 5
ACCESS IC LAB
Graduate Institute of Electronics Engineering, NTU
Pipelining
original
Tmax
fclock < 1/T max
data_in
Input
Registers
Output
Registers
Multi-level Logic
data_out
2-stage pipelining
data_in
data_out
Input
Registers
Comb
Logic
Pipeline
Register
Comb
Logic
Output
Registers
fpipeline < 2 / T max
Tmax/ 2
P. 6
ACCESS IC LAB
Graduate Institute of Electronics Engineering, NTU
Example 1 : Simple Circuit
original
3-stage pipelining
Critical path = 3 adders
Critical path = 1 adders
P. 7
ACCESS IC LAB
Graduate Institute of Electronics Engineering, NTU
Example 2 : Pipelined 16-bit Adder
clock
b[15:0]
a[15:0]
a[15:8]
b[15:8]
b[7:0]
a[7:0]
IR[16:9]
b
IR[32:25]
IR[24:17]
c_in
IR[8:1]
Input Register: IR[32:0]
IR[0]
a
c_out
c_in
sum
Pipeline Register: PR[7:0]
PR[24:17]
b
PR[16:9]
a
c_out
c_in
PR[8]
PR[15:0]
sum
sum[7:0]
c_out
Output Register: OR[16:0]
sum[15:0]
P. 8
ACCESS IC LAB
Graduate Institute of Electronics Engineering, NTU
Parallel Processing
original
Parallel processing
P. 9
ACCESS IC LAB
Graduate Institute of Electronics Engineering, NTU
Area Issues
Area = Cost.
During the design process, the designer should
be “area aware”.
Resource sharing
P. 10
ACCESS IC LAB
Graduate Institute of Electronics Engineering, NTU
Power Consumption in CMOS
Low power design is more and more
important in today’s chip design due to heat
dissipation, packaging, and portability needs.
Power Consumption
P Nnode * CL *Vdd * fclock
2
Nnode : switching activity
fclock : clock frequency
CL : node capacitance
Vdd : power supply voltage
P. 11
ACCESS IC LAB
Graduate Institute of Electronics Engineering, NTU
Strategy for Low-Power Design (1/2)
Vdd is technology-dependent.
Pipelining
Parallel processing
Execute tasks concurrently to improve throughput
Increase the system’s overall sampling rate
Incorporates multiple copies of hardware
CL can only be minimized by back-end design.
Optimize fclock and Nnode are the most practical
power reduction techniques.
P. 12
ACCESS IC LAB
Graduate Institute of Electronics Engineering, NTU
Strategy for Low-Power Design (2/2)
Reducing Clock Frequency
Design with clock rate that is ‘just right’
Clock Gating
Reducing switching activity
Avoid unnecessary circuit switching
Reducing switching activity at I/O pins
Use simple hardware if it gets the job done
P. 13
ACCESS IC LAB
Graduate Institute of Electronics Engineering, NTU
Example : Multiplier
original
Low-power design
P. 14