Non – overlapping Clock Generator

By: Hem Doshi

Outline

• Introduction• Architecture

– Working functionality– Why?

• Test Logic Implementation• Conclusion

– Based on our need, and most proven on Silicon!

Conceptual circuit and Signal Flow

Del

ay

Delay Equations

Del

ay

Architecture

Highlights

• Conventional Design• Proven on Silicon• Usage of minimum size NAND and then scaling

Super Buffer to drive larger load.• Usage of Inverters to attain Non-Overlap.• # of Inverters in feedback controls Non-Ov-Time • Advanced clocks can be tapped out a couple

Inverters before delayed clock.• Easy knobs available to control Tnov and Tlag

Design Flow• Calculating Load Capacitance

– Major Contributing Factors• Gate Caps of all the transistors driven by clock

– Stage 1 through 6– Main Sample/Hold Circuit– Comparator of final stage… ?– Digital Decode buffers

• Parasitics– clock distribution network

» Fringe Component» Area Component

– What I haven’t consider• Drain Cap of PMOS/NMOS of final stage of inverter chain

– As it scales with the inverter chain itself– Have some margin and design to drive extra load

• Usage of minimum L– Realization of sharp edges – circumvents jitter @ distribution

• Total ‘W’ for final stage = Cload/(Lmin*Cox* 3)

Design Flow (Continued)• Distribute total W with 3:1 ratio between PMOS and

NMOS– Ensures symmetric rise and fall time– Assumes mobility ratio of 3:1 between Pmos and Nmos

• Continue shrinking W by factor of 3 until hit by minimum W supported by process

• Calculate the delay incurred by the driver chain• If required Non-Ov-time is greater ~ 4.9ns

– add series of minimum size inverters in the feedback chain to obtain required delay

• Advanced clock driver is some smaller then Delayed clock as load Cap is diff.

Total W for Stage 1

Total W for Stage 2

Total W for Stage 1-6

• Ø1 = 3 * (220+88) = 924 W• Ø2 = 3 * (220+88) = 924 W• Ø1’= 3 * (72) = 216 W• Ø2’= 3 * (72) = 216 W

Total W for S/H Circuit

Digital Decode and 7th stage

Parasitics

• Active Chip Area is 1×1 mm2

• Estimated Routing channel length in Metal 3 is– 1 mm + 1 mm + 0.4 mm = 2.4mm

• Metal 3 to Metal 3 capacitance Calculation– Total distance between Metal 3 and Substrate is

2.575 um– Distance between Clock Distribution channels is 2

µm..!– Not a huge difference, so a good estimate would be

to associate total fringe Capacitance to either Metal to metal or metal to Substrate Fringe.

Total Fringe and Area Component

Total Parasitic Capacitance is thus 141.12 + 350.4 = 491.42fF ≈ 500fF ≈ 0.5pF

Final Cap Calculation

Test Logic

• Implements Tri-state buffers to reduce/increase the delay in feedback path.

• Lay out inverter chains with different delay– Preliminary Desgin

• 3 chains (1 ns, 2 ns, 3 ns) delay

• Last inverter in each series will have ability to tri-state its output

• Succeeding chain will bypass the last inverter in preceding chain

• Test logic chains will have odd # of inverters.

Tri State Buffer

Proposed Test Logic Model

Test Logic model (Continued)

100011

010001

001010

000100

Chain4Chain3Chain2Chain1Sel(2)Sel(1)

OutputTest Input

Truth Table

Timing Calculations

• Worst Case OP-AMP Settling time Calculation:• Worst case SS corner UGBW for the OP-AMP is

250 MHz.• Worst-Case settling time• = = 17.825 nS• Worst Case Comparator Settling Time ≈ 3 ns• Total Worst Case time

• 3 + 17.825 = 20.825ns ≈ 21 ns

• Available max duty cycle is 50% of 20 MHz clock. Thus max time available is 25 nS.

FactorBFUGBW ..27××× π 4

12102507

6 ×××× π

Timing Calc (Contd.)

• Available theoretical margin = 25 – 21 = 4nS.• Equally dividing for between OPAMP and

Comparator settling makes on-time of clock = 20 nS.

• Required duty Cycle = 20/50 =40%• T-T corner Non-Overlap time is 5nS and

available margin would be ±2nS.

Process Corners

0.2970.2584.431.45TT27 temp

5 V

0.1670.1672.7850.991FF0 temp5.5 V

0.4240.3966.962.39SS85 temp

4.5 V

TfTrTnovTlag

Actual Supply

Testbench

Sizing of the Device

Testbench

Fall Time

Rise Time

Tlag

Tnov

Bad Wave !