Summary of lecture videos provided by anonymous online watcher - Contribution is gratefully acknowledged.
1.EE141 - 1/20/2012 - Introduction. Introduction to the course and Evolution of Computers.
2.EE141 - 1/25/2012 - Fabrication, Yield, Design Metrices - Cost and Reliability(Noise and Sources of Noise), Introduction to some terms related to VTC
3.EE141 - 1/27/2012 - Regenerative property,Design Metrices- Performance (Delay Definitions and first order transient analysis) and Power, Energy and Power Delay Product,Transistor Layout and Design Rules
4.EE141 - 2/1/2012 - Design rules, Switch logic. Switch model of MOS and CMOS inverter ,CMOS inverter (first order DC analysis),Symmetrical VTC,CMOS inverter transient response,Sizing of Inverter Chain*(Performance)
5.EE141 - 2/3/2012 - Switch logic. Inverter Chain Sizing Continued and its impact on energy and area; nmos(strong 0) and pmos(strong 1); CMOS logic
6.EE141 - 2/8/2012 - Complex logic. Sizing in complex logic,Delay(Sizing) of NAND and NOR,Why is NAND faster than NOR in CMOS logic(logical effort),Logical Efforts of various gates,How to decide which circuit is better for performance(Load small then intrinsic delay factor(p) dominates and Large Load Logical Effort is deciding factor)
7.EE141 - 2/10/2012 - Wires. Branching Effort Examples,Wires, Varying (increasing)Thickness of Wires as we go up the layers, Wire Models, Parallel plate capacitor model, Fringing Capacitance, Capacitative coupling and noise, Resistance Of wires
8.EE141 - 2/15/2012 - Wires + logic. Silicides reduces resistance,lumped model, Resistance of transistor is of order 10^3ohm, Distributed RC model(Delay is only half of that predicted by lumped RC model ), Elmore Delay ,Diffusion of step signal through distributed rc leads to less steeper exponentials as we move towards destination,Insertion of Repeaters in RC n/w,Decoders and optimization
9.EE141 - 2/22/2012 - Memory Decoder (continued), Predecoders, Physics of MOS,Threshold Voltage,Body Effect, Linear and Saturation Region, Pinch Off,Drain Current Equations,Channel Length Modulation, Velocity Saturation, Short Channel device saturates early,Unified Model,Approximating Resistance of Transistor
10.EE141 - 2/24/2012 - Transistor capacitances. DIBL, MOS capacitances, Overlap capacitances, Channel Capacitance in different regions of operation,Diffusion(Junction) Capacitance, Channel Stop Implant,Linearizing the junction capacitance, Derivation of VTC of CMOS inverter
11.EE141 - 2/29/2012 - For Vm=VDD/2 Rn=Rp,Strong nmos means wider nmos(or Rn is low),Linear approximation of VTC, VTC for varying VDD, Inverter characterstics below threshold(subthreshold),VDD>50mv for inverter characterstics, MID Term Solutions, Discussion of Project
12.EE141 - 3/2/2012 - Adders + Multipliers. Discussion of project(continued), Ripple Carry Adder(Tdelay is O(N)),Direct Implementation(28 transistors),Mirror Adder(Improve the logical effort of A and B inputs), Sizing the mirror adder, Keep the critical inputs near the output,Carry Bypass(Carry skip) Adder(O(N) but Td proportional to N/4), Carry Select Adder(O(N)), Increase Block bit length to make square root carry select(O(sqrt(N))), Rearrange the block bit length (increase then decrease) to get a faster carry bypass adder
13.EE141 - 3/7/2012 - Arithmetic (missing video - slides only) - Carry Look Ahead adders, Logarithmic (Tree) Adders(Best adder available O(log2N)),Brent Kung Tree(Disadv - Large Fanout), Kogge Stone (Disadv - Large area , Adv- Fastest adder),Binary redundance representation,Relational operators implementation,Multipliers, Array Multiplier,Carry Save Multiplier,Wallace Tree Multiplier(log1.5N),Booth Multiplier
14.EE141 - 3/9/2012 - Memory. Memories, Static Memory(Data bit stored as voltage), Dynamic Memory (Data bit stores as charge), Memory Classification, 2 stable and 1 meta stable states in two inverter SRAM cell,Writing into cross -coupled pair inverter (overpower the feedback),Ratioed Logic, Always write "0"(get ratio b/w access transistor((nmos) and pmos,make nmos(access) strong than pmos), From read (get ratio b/w access transistor(nmos) and nmos,make acess transistor(nmos) weak than nmos ), Read static noise margin
15.EE141 - 3/14/2012 - Memory. SRAM ( 6-T Contd), Resistive Pull up SRAM, DRAM, 3-T DRAM cell( needs to be refreshed, inverted value is read,dedicated bit line for read and write(due to leakage but read is non destrcutive), 1-T DRAM cell(Destructive read thus sense amplifier read and restore data), CMOS switching delay, Approximating the non linear resistance of transistor(1.Current Source(IdSAT),2. Current Source including some finite resistance 3.Resistor Approximation 4.From 2 Find Req), Resistance variation with Vdd, Capacitances(Contribution of various capacitances in load capacitance)
16.EE141 - 3/16/2012 - Power Dissipation in CMOS (1. Switching Power(Dynamic Power)(capacitor charge) 2. Leakage(subthreshold, Reverse Leakage Current) 3.Short Circuit Power 4.Bias Power), Capacitance values are different in power analysis from delay analysis (as we go from 0->Vdd rather than 0->Vdd/2),Make input rise time equivalent to output fall time to minimise short ckt current, To reduce subthreshold current try to reduce the slope below Vth(80-90 mv reduction in voltage for decade reduction in current), Threshold variation with Vds(DIBL) and Length L, Design Techniques for performance : 1.Fan in Considerations in Delay(Tapered Sizing),Gates with fan -in greater than 4 must be avoided(performance), Delay increases linearly with Fan Out but quadratically with Fan In; 2. Transistor reordering (Keep critical inputs closer to output); 3. Alternate ckt implementation; 4. Isolate Fan in and Fan out; 5. Reduce Voltage Swing.
17.EE141 - 3/21/2012 - Pass-transistor and ratioed logic. Moving to Ratioed Logic (Performance),Trade off is Power,prefer NOR gates in nmos PDN ratioed logic(reason-NAND series nmos),For Calculating LE find the active paths to VDD/Ground from output then take RgCg/RinvCinv,Conflicting Currents take Reff=1/(1/Rn-1/Rp),Improved loads - 1. Adaptive loads( In PUN Two transistors , one with large W and one with small W . While transistion use larger W),2.Differential Cascode Switch Voltage Logic (DCSVL)(Not ratioed anymore and no static power dissipation), Pass Transistor Logic,NMOS only logic and Level Restorer(Adv- Full Swing and no static power dissipation in inverter disadv-ratioed logic between pass transistorand Level restorer pmos),Complementary PTL(Instead of using level restorer use differential output ), Transmission gates(added pmos thus slower), Transmission gate multiplexers and XOR gates, Dynamic Logic (Adv over ratioed logic full swing , non ratioed, no static power more dynamic power (more transistion and more clk distrbution),faster than CMOS)
18.EE141 - 4/4/2012 - Dynamic logic. Mid term 2, Dynamic Logic( Faster than CMOS and also non-ratioed,no static dissipation),NOR gates are better in dynamic logic style than NAND gates,Dynamic inverter has very small Noise Margin, Challenges in Dynamic Logic - 1.Charge leakage(rev biased diode) (Solution - 1.operate at high freq and dont give time to discharge 2. Bleeder or Level restorer(inverter feedback)) 2.Charge Sharing
19.EE141 - 4/6/2012 - Layout. 2.Charge Sharing( Solution - 1.Bleeder 2.Pre charge intermediate nodes (bleeder)),3.Clock Feedthrough (Bootstrapping)(Diode p-n in pmos might become FB and might lead to Latch Up) (Solution is having substrate contects to provide currents to go to ground or Vdd),4.Backstage Coupling , Cascading of Dynamic Inverters must be done through Domino Logic(only 1->1 or 1->0 transistion allowed before second dynamic invereter) (soln. 1(Domino logic)add static inverter in b/w also add level restorer, 2. Delayed Clocks), Dynamic logic is non-inverting, NOR gates better than NAND in Dynamic(Domino) Logic ,Assymetrical VTC(for the static inverter just after the dynamic gate) to reduce tpLH makes it faster(reduce logical effort),can eliminate evaluate transistors(except first stage) in Domino logic
20.EE141 - 4/11/2012 - Sequential circuits. -CMOS layout, Euler Paths for efficient layout(continuous diffusion stripes for both pmos and nmos),, Multifingered Transistors, Sequential Logic, Latches vs Flip Flops,Set up and Hold Times,Writing into static latch(using MUX)
22.EE141 - 4/18/2012 - Clock skew. (at two different places),its effect on inequalities,Clock jitter(at the same place) and its effect on inequalities,Jitter has Gaussian Distribution,Sources of clock uncertainity,Clk distribution should be such that it should reach different points of ckts at same time,Latch based clocking,Time borrowing in latches,Flip flop or Register(adv is no race conditiob but disadv is not lenient with clock shortcomings(Hard edges)), Latch(adv is leniency with clock shortcomings,Time Borrowing but disadv is race condition), Clock distribution(H-Tree), clock grid(done locally only as consumes huge power)
23.EE141 - 4/20/2012 - Clocks and Power. What matters is to minimise clock skew and not clock delay,PLL is used to multiply the frequency of crystal clock,Various clocking systems development,Power Distribution,inverter is not a good power supply noise rejection device,Multiple pins for power distribution,layers for vdd and ground,Electromigration( soln -Limit the current per unit length)(electrons push the ions and thus they move to different positions and can cause short or open ckts),Add decoupling capacitance(Gate capacitance is used) to avoid sudden drop in v of supply rails,Rc n/w do not produce ringing oscillations,Real Power distribution analogy is now coming to new generations chips,switches in power line, when a block is not in use turn off clock as well as power supply,Technology scaling,Full scaling(constant EF),Fixed voltage and general scaling
24.EE141 - 4/25/2012 - Scaling + Energy. Vertical channel wrapped with gate provides more control over channel(FINFET Or Trigate),Carbon Nano Tubes (Cylindrical covering of channel with gate),Wire Scaling,Transition activity and power,Low Power Design principles,Energy Performance Curve,Concurrency(Reducing clk freq(same performance) and Area for low power),Parallelism(How many processors in parallel depends majorly on reduction of VDD) and Pipelining(we should scale Vdd utilizing the speed up obtained if we want low power design),Time Multiplexing to use same block again and again
25.EE141 - 4/27/2012 - Perspectives. Leakage Power, If a block is not being used turn off the clock as well as disconnect from data bus,Staic power (Leakage power) reduction - 1.Use switches(Power gating) to disconnect power when not in use, 2.Body biasing - increase the Vth when not in use so that leakage current reduces substantially, 3.Supply voltage Ramping->ramp from Vdd to 0 when not in use 4. Transistor stacking(Increase resistance and for upper transistor Vth also increases due to body effect and thus current reduces more),Power Gating(implicit transistor stacking) MTCMOS(Multi threshold CMOS), Digital Circuit - Future Possibilities