Download - Adiabatic Logic
![Page 1: Adiabatic Logic](https://reader034.vdocuments.us/reader034/viewer/2022042511/55cf98ad550346d03399098e/html5/thumbnails/1.jpg)
Adiabatic Logic
Yasmine Badr
NanoCAD Lab
1
A major part of the contents is from Reference 1
![Page 2: Adiabatic Logic](https://reader034.vdocuments.us/reader034/viewer/2022042511/55cf98ad550346d03399098e/html5/thumbnails/2.jpg)
Terminology
• Reversible logic:
– circuits that have one-to-one mapping between vectors of inputs and outputs
• “Adiabatic” : from thermodynamics
– Process in which there is no exchange of heat with the environment
2
![Page 3: Adiabatic Logic](https://reader034.vdocuments.us/reader034/viewer/2022042511/55cf98ad550346d03399098e/html5/thumbnails/3.jpg)
History
• Landauer (1961):
– Logically irreversible operations in a physical computer necessarily dissipate energy
– Logically reversible operations “can” be performed without dissipation
• Later, found that that a gate can work energetically reversible without the need to be logically reversible.
3
![Page 4: Adiabatic Logic](https://reader034.vdocuments.us/reader034/viewer/2022042511/55cf98ad550346d03399098e/html5/thumbnails/4.jpg)
Adiabetic Circuits
• Energy reused rather than just dissipated
• Power advantage
4
![Page 5: Adiabatic Logic](https://reader034.vdocuments.us/reader034/viewer/2022042511/55cf98ad550346d03399098e/html5/thumbnails/5.jpg)
Conventional CMOS
• Constant voltage • Energy dissipated
per transition =CVdd
2/2 – Pull up: energy
dissipated on PMOS – Pull down: charge
dissipated via NMOS to ground
• Energy used only once
5
![Page 6: Adiabatic Logic](https://reader034.vdocuments.us/reader034/viewer/2022042511/55cf98ad550346d03399098e/html5/thumbnails/6.jpg)
Adiabatic Logic
• Supply is a power clock (ramp)
• Pull down path: to power supply
6
![Page 7: Adiabatic Logic](https://reader034.vdocuments.us/reader034/viewer/2022042511/55cf98ad550346d03399098e/html5/thumbnails/7.jpg)
Adiabatic Logic Energy Dissipation
i(t)=c dV/dt
=c Vdd/T
(Constant Current)
Energy during transition time T
= I2*R*T
= R* C2 * Vdd2/T
7
Equivalent circuit
![Page 8: Adiabatic Logic](https://reader034.vdocuments.us/reader034/viewer/2022042511/55cf98ad550346d03399098e/html5/thumbnails/8.jpg)
CMOS
• Dissipated energy in pull up and pull down
= CVdd2
• Depends on C and Vdd only
Adiabatic
• Dissipated energy in charge and recovery
= 2*R* C2 * Vdd2/T
• Depends on size of transistor and T as well
• Slower circuit is charged less energy dissipation
8
T>2RC Adiabatic has less energy dissipation if we assume activity factor of 1
![Page 9: Adiabatic Logic](https://reader034.vdocuments.us/reader034/viewer/2022042511/55cf98ad550346d03399098e/html5/thumbnails/9.jpg)
Adiabatic Logic Families
• Adiabatic system has:
– Digital core: adiabatic gates
– Generator of power clock signals
• 4-phase power clock for cascaded gates
• Efficient generation essential for high energy saving factor
• Two of the most popular
– Efficient Charge Recovery Logic (ECRL)
– Positive Feedback Adiabatic Logic (PFAL)
9
![Page 10: Adiabatic Logic](https://reader034.vdocuments.us/reader034/viewer/2022042511/55cf98ad550346d03399098e/html5/thumbnails/10.jpg)
1-ECRL Inverter
• Logic implemented using NMOS
• 2 cross coupled PMOS to store info
• For a different function e.g. NAND – a series connection of 2 NMOS
instead of NF, using A and B as input s.
– 2 parallel transistors, having A’ and B’ as inputs instead of NF’
10
![Page 11: Adiabatic Logic](https://reader034.vdocuments.us/reader034/viewer/2022042511/55cf98ad550346d03399098e/html5/thumbnails/11.jpg)
1-ECRL Buffer/Inverter (cont’d)
• Input signal is shifted by 90 degrees w.r.t to the power clock
11
![Page 12: Adiabatic Logic](https://reader034.vdocuments.us/reader034/viewer/2022042511/55cf98ad550346d03399098e/html5/thumbnails/12.jpg)
1-ECRL Buffer/Inverter (cont’d)
• Power clock ramping up from 0 to Vdd
• Out follows Power clock
12
1
0
0
>Vthp
φ
0
![Page 13: Adiabatic Logic](https://reader034.vdocuments.us/reader034/viewer/2022042511/55cf98ad550346d03399098e/html5/thumbnails/13.jpg)
1- ECRL Buffer/Inverter (cont’d)
• Power clock ramping down from Vdd to zero
• Both inputs are now zero – Because preceding
stage is now recovering energy
– Not strictly complementary
• N1 and N2 off
13
0
0
0
>Vthp
0
Discharge path
![Page 14: Adiabatic Logic](https://reader034.vdocuments.us/reader034/viewer/2022042511/55cf98ad550346d03399098e/html5/thumbnails/14.jpg)
1- ECRL Buffer/Inverter (cont’d)
• Φ < Vth,p
• Fraction of energy
= ½ CoutV2
th,p
dissipated or reused next cycle according to succeeding input signal
Quasi-adiabatic
14
0
0
0
<Vthp
0
Vthp
![Page 15: Adiabatic Logic](https://reader034.vdocuments.us/reader034/viewer/2022042511/55cf98ad550346d03399098e/html5/thumbnails/15.jpg)
2-PFAL Inverter/buffer
• Consists of:
– Latch made of 2 cross-coupled inverters
– Logic function
15
![Page 16: Adiabatic Logic](https://reader034.vdocuments.us/reader034/viewer/2022042511/55cf98ad550346d03399098e/html5/thumbnails/16.jpg)
ERCL vs PFAL
• ERCL: less number of transistors
• PFAL:
– Functional block parallel to PMOS
less equivalent resistance
less energy dissipated (R* C2 * Vdd2/T)
16
![Page 17: Adiabatic Logic](https://reader034.vdocuments.us/reader034/viewer/2022042511/55cf98ad550346d03399098e/html5/thumbnails/17.jpg)
4-Phase Power clock for Cascade
• More than 1 power clock used to operate adiabatic system – ERCL and PFAL use 4-phase
power clock φ0-φ3
• Because input has to be stable in evaluation phase – ERCL and PFAL: 90° phase
shift between subsequent phases is obtained
17
![Page 18: Adiabatic Logic](https://reader034.vdocuments.us/reader034/viewer/2022042511/55cf98ad550346d03399098e/html5/thumbnails/18.jpg)
4-Phase Power clock for Cascade (cont’d)
• 4 – Phases:
– Evaluate (E): outputs are evaluated from stable input signals
– Hold (H): outputs are kept stable to supply subsequent gate with stable input signal.
– Recover (R): Energy is recovered
– Wait (W): for symmetry reasons because symmetric signals are easier to generate
18
![Page 19: Adiabatic Logic](https://reader034.vdocuments.us/reader034/viewer/2022042511/55cf98ad550346d03399098e/html5/thumbnails/19.jpg)
Power Supply
• LC circuit
• Stepwise charging
– Charging of output to VDD is not done abruptly, but is divided into N steps
19
![Page 20: Adiabatic Logic](https://reader034.vdocuments.us/reader034/viewer/2022042511/55cf98ad550346d03399098e/html5/thumbnails/20.jpg)
Cons and Pros
- Slower than conventional CMOS
- Requires special power supply
- Area (but we get function and its complement)
+ Less Power if high switching activity
or disconnect system from power supply while idle (sleep transistors)
20
![Page 21: Adiabatic Logic](https://reader034.vdocuments.us/reader034/viewer/2022042511/55cf98ad550346d03399098e/html5/thumbnails/21.jpg)
References
1. Teichmann, Philip; Adiabatic Logic; Springer; 2012
2. Vitányi, P.; Time, space, and energy in reversible computing; Proceedings of the 2nd conference on Computing frontiers, ACM, 2005, 435-444
3. Pahlavan,B; Evaluation of Trends in Adiabatic Logic for Low Power Design http://cutler.eecs.berkeley.edu/classes/icdesign/ee241_s06/projects/midterm/pahlavanghanadanskucha.pdf
4. Frank, M; Reversible Computing and Truly Adiabatic Circuits: Truly Adiabatic Circuits: The Next Great Challenge for Digital Engineering; 2006
5. Sanjay Kumar; Design Of Low Power Cmos Cell Structures Based On Adiabatic Switching Principle; Master Thesis; 2009
21