Reading Assignment: Rabaey: Chapter 7 ELEC 516 VLSI System Design and Design Automation Spring 2010 Lecture 5: Flip-Flop/Latch Design Reading Assignment: Rabaey: Chapter 7 Note: some of the figures in this slide set are adapted from the slide set of “ Digital Integrated Circuits” by Rabaey et. al., Copyright 2002

Motivations: Why do we need sequential circuits? Need memory Pipeline the system so that new operations start before the old ones complete. Add registers to keep operations separate. Convert parallel operations to a sequence of serial operations (faster operations per cycle/ smaller). Need to process a sequence of inputs and want to reuse the same hardware (Finite State Machine)

Sequential Logic Memory Element: F s LOGIC t p,comb f In Out Memory Element: Stores a value as controlled by clock. May have load signal, etc 2 storage mechanisms positive feedback charge-based

Memory Elements - Latches and flip-flops A generic memory element has an internal memory and some circuitry to control access to the internal memory. which is controlled by the clock input. Memory element differ in many key respect: exactly what form of clock signal causes the input data value to be read; how the behavior of data around the read signal from clock affects the stored value; when the stored value is presented to the output; whether there is a combinational path from the input to the output. 2 types of memory: latches and edge-triggered flip flop latches - transparent while the internal memory is being set from the data input edge-triggered flip-flops (or register) - not transparent, reading the input value and changing the flip-flop’s output are two separate events.

Simple Circuit with Feedback One inverter with feedback Self-oscillation, 2 gate delays for one period Odd-number of inverters with feedback self-oscillation of 2x gate delays of one path Two inverters with feedback Memory element (or states) Basis for commercial static RAM designs Read-only, but has no write function Memory with read/write capability Selectively break the feedback path by transmission gates to load new value into the cell A can be written to Z when LD = 1 Write SW On & Feedback SW Off Z holds the value when LD = 0 Write SW Off & Feedback SW On “0?" "1" "0" "1" Z LD LD’ A

Latch versus Register Latch stores data when clock is low D Q D Q Clk stores data when clock rises D Q D Q Clk Clk Clk Clk D D Q Q

Let’s build a latch Latches are multiplexers controlled by a clock: When CLK is high data will pass through otherwise the data is saved or kept unchanged feedback from output to input Can be realized using transmission gates

Dynamic latch Stores charge on inverter gate capacitance: Uses complementary transmission gate to ensure that storage node is always strongly driven. Latch is transparent when transmission gate is closed. Storage capacitance comes primarily from inverter gate capacitance. Setup and hold times determined by transmission gate—must ensure that value stored on transmission gate is solid.

Dynamic latch- Stored charge leakage Stored charge leaks away due to reverse-bias leakage current. Stored value is good for about 1 ms. Value must be rewritten to be valid. If not loaded every cycle, must ensure that latch is loaded often enough to keep data valid.

Non-dynamic latches Must use feedback to restore value. Some latches are static on one phase (pseudo-static)—load on one phase, activate feedback on other phase. Example - Recirculating latch: Static on one phase:

Latch-Based Design N latch is transparent when f = 0 P latch is transparent when f = 1 f N P Logic Latch Latch Logic

Positive Feedback: Bi-Stability The circuit presents only three operation points When the gain of the inverter in the transient region is larger than 1, A and B are the only stable operation points & C is a metastable operation point.

Meta-Stability Gain should be larger than 1 in the transition region C is an instable operating point. Every deviation (even small) causes the operation to run away (because of high gain). A and B are very stable operation points, the loop gain is much smaller than unity, eve a large deviation will not cause deviation from these operation points.

Flip states in Bistable Circuit Two different approaches: Cutting the feedback loop Open the loop and write data Multiplexer based Q = Clk’.Q + Clk.In Overpowering the feedback loop Applying a trigger signal at the input of the flip-flop to overpower the stored value to a new value Careful sizing of the transistors in the feedback loop and the input is necessary

Writing into a Static Latch Use the clock as a decoupling signal, that distinguishes between the transparent and opaque states D CLK Forcing the state (can implement as NMOS-only) Converting into a MUX

Other styles Small and lower clock load but sizing problems Very good style (Skew considerations) Fast and energy efficient Presents the lowest clock load

Mux-Based Latch Positive level latch. When the D path is ON, the feedback is cut-off No sizing issues for correct operation. The number of transistors that the clock drives is an issue (clock has an activity factor of 1: CLK Load of four transistors.

Mux-Based Latch NMOS only Non-overlapping clocks

DFF Implementation (falling edge triggered) Master/Slave latch arrangement Ds D Q D Q D Q G Q’ G Q’ Cs C D Q Master D latch Slave D latch C Q’

DFF Internal Operation C Master sampling Master sampling Ds Xfer to Slave Cs Xfer to Slave Q

Flip-Flop: Timing Definitions Setup time: time before clock during which data input must be stable. Hold time: time after clock event for which data input must remain stable. Clock-to-Q delay = TPFF

Clk-Q Delay

The setup time race Setup represents the race for new data to propagate around the feedback loop before clock closed the input gate. If data arrives too close to clock edge, it will not set up the feedback loop before clock closed the input TG

The hold time race Hold time represents the race for clock to close the input gate before next cycle’s data disturbs the stored value If data changes too soon after the clock edge, clock might not had time to switch off the input gate and new data will corrupt feedback loop

Setup Time

Maximum Clock Frequency Also: tcdreg + tcdlogic > thold tcd: contamination delay = minimum delay tclk-Q + tp,comb + tsetup = T Modern high performance systems are characterized by low logic depth The register’s delay becomes very important as the registers because it accounts for both the setup and propagation delay. DEC Alpha up has a max logic depth of 12 gates, &15% of the delay corresponds to the register overhead.

Reduced Clock Load Master-Slave Register Eliminate the feedback transmission gate. Possible problem: reverse conduction

Clock Overlap Problem Potential problems: X CLK CLK Q A D B (a) Schematic diagram CLK CLK Potential problems: Race condition: data at the output change at the rising edge of the clock Node A can be driven by both D and B when clock overlap CLK CLK (b) Overlapping clock pairs Variations in the routing wires Used to route CLK & CLK’ Variations on the load Inverter’s delay

2 phase Non-overlapping clock DFF Two-phase non-overlapping clocks

6 Transistor CMOS SR-Flip Flop

Dynamic Latches and registers Disadvantage of static FF - complexity, larger size The requirement that the memory should hold state for extended periods of time can be relaxed in computational structure Dynamic - use the charge stored in capacitance to eliminate the use of inverter pair to latch data: Pseudo-static latch (Charge-Based Storage) A D Clk Clk’ B Q Clk=1: The input data is sampled on storage node A, during this time the slave is on hold mode, node B at high impedance. On the falling edge of the clock, T2 turns ON and the value sampled on A propagates to the output. Setup time is the delay of the transmission gate Hold time is zero since the TG is turned off on the clock edge. Tc2q= delay of two inverters and a transmission gate.

Making a Dynamic Latch Pseudo-Static Fully dynamic circuit presents a number of drawbacks: Capacitive coupling can inject significant noise to the internal storage node. Leakage current: Most modern processors require that the clock can be slowed down or completely halted to conserve power in low activity periods Most of these problems can be addressed by adding a weak feedback inverter  Pseudo-Static Latch Slight cost in delay and silicon area. Improves noise immunity significantly Dynamic latches should be made Pseudo-Static, keep for very special cases of highly controlled environment (full-custom, high performance data path design)

Impact of non-overlapping clocks D Clk Clk’ B Q T1 T2 clk During the (0,0) overlap, PMOS of T1 and PMOS of T2 are ON, creating a direct path from D to Q. The same problem appears during the (1,1) overlap. clk (0,0) Overlap (1,1) Overlap Overlapping Clocks Can Cause Race Conditions Undefined Signals

Flip-flop insensitive to clock overlap DD DD M 2 M 6 f M 4 f M 8 X In D C C f M 3 L 1 f M 7 L 2 M 1 M 5 f- section f - section C 2 MOS LATCH

C2MOS avoids Race Conditions In 1 M 3 2 6 7 5 V DD 4 8 (a) (1-1) overlap (b) (0-0) overlap X During the (0,0) overlap, new data sampled on the falling edge of the clock will not appear at D output. Same remark applied during the (1,1) overlap.

Operation of the C2MOS latch When f = 1, M3 and M4 are on, the 1st section is in the evaluation mode and the second section is in a hold mode (high impedance). M7 and M8 are off, decoupling the output from the input. The input D retains its previous value stored on the output capacitance. When f = 0, the first section is in hold mode and the second section is in evaluation mode, the value stored in CL1 propagates to the output node. A C2MOS register with (f - f ) clocking is insensitive to overlap, as long as rise and fall times of the clock edges are sufficiently small.

Dual Edge registers Dual Edge registers are very interesting as they permit to run the Clock 2 times slower  lower power on the clock node. D f N1 X N2 Q Y

Pipelined Logic using C2MOS F Out f V DD f V DD NORA CMOS G C C C 1 2 3 What are the constraints on F and G? No-race rule: A C2MOS-based pipelined circuit is race-free as long as the logic function F (static logic) between the latches are non-inverting

Example f V DD V DD f 1 f Number of a static inversions should be even

Doubled C2MOS Latches-True single-phase clock register f V DD Out V V DD DD In In f f Out Doubled n-C2 MOS latch (transparent when CLK= 1) Doubled p-C2 MOS latch (transparent when CLK= 0) + Requires a single clock to build a positive & negative clock - Can suffer from charge sharing & noise pbs when clk low.

TSPC - True Single Phase Clock Logic f V DD In Static Logic PUN PDN f V DD Out Including logic into Inserting logic between the latch latches

Example of Including Logic in TSPC CLK V DD Q In 1 2 Embedding logic into the latch reduces the delay overhead of the latch. This approach of embedding logic into the latch was extensively used in many high performance microprocessors including EV4 DEC Alpha. AND latch

Master-Slave Flip-flops DD D f V DD D f V DD D f V DD D Y X D (a) Positive edge-triggered D flip-flop (b) Negative edge-triggered D flip flop (c) Positive edge-triggered D flip-flop using split-output latches

Pulse-Triggered Latches An Alternative Approach Ways to design an edge-triggered sequential cell: IDEA: construct a short pulse around the rising (or falling) edge of the clock. This would be the NEW clock input. Hold time is equal to the length of the pulse. Master-Slave Latches Pulse-Triggered Latch L1 L2 L Data Data D Q D Q D Q Clk Clk Clk Clk Clk + Reduced clock load, small number of transistors required. + Glitch circuitry can be shared by multiple register. - increase in verification complexity: Need to make sure the glitch Is properly generated!!!!!

Pulsed Latches Glitch corresponds to the delay of the AND + 2 Inv

Pulsed Latches Hybrid Latch – Flip-flop (HLFF), AMD K-6 and K-7 : When the clock is low: M3 and M6 are off and P1 is ON. Node X is precharged to VDD, and the output is decoupled from X (Memory). CLKD’ is a delayed-inverted version of CLK. On the rising edge of the clock, M3 and M6 turn ON while M1 and M4 stay ON for a short period, and hence the latch is transparent and D is sampled. Once CLKD’ goes low node X is decoupled from D.

Hybrid Latch-FF Timing Advantage: Setup time can be negative: Transparency window Is longer than the delay from input to the output. D injected after the clock.

Pipelining REG LOG a b Out f REG a b f LOG REG Out f REG f REG f

Latch-Based Pipeline

NORA (NO Race) CMOS Modules f block Logic Latch f =0 f =1 Precharge Hold Evaluate

Non-Bistable Sequential Circuits─ Schmitt Trigger VTC with hysteresis Restores signal slopes

Noise Suppression using Schmitt Trigger

CMOS Schmitt Trigger Moves switching threshold of the first inverter

Schmitt Trigger Simulated VTC 2.5 2.5 2.0 2.0 V 1.5 M 1 1.5 (V) (V) X x 1.0 V 1.0 V M 2 V k = 1 k = 3 k = 2 0.5 0.5 k = 4 0.0 0.0 0.0 0.5 1.0 1.5 2.0 2.5 0.0 0.5 1.0 1.5 2.0 2.5 V (V) V (V) in in Voltage-transfer characteristics with hysteresis. The effect of varying the ratio of the PMOS device M . The width is k * 0.5 m. m 4

CMOS Schmitt Trigger (2)

Multivibrator Circuits

Transition-Triggered Monostable

Monostable Trigger (RC-based)

Astable Multivibrators (Oscillators) 1 2 N-1 Ring Oscillator simulated response of 5-stage oscillator

Voltage Controller Oscillator (VCO)

Differential Delay Element and VCO ctrl o 2 1 in delay cell v 1 2 3 4 in 2 two stage VCO 0.5 0.0 1.0 1.5 2.0 2.5 3.0 2 V 1 3 4 time (ns) 3.5 simulated waveforms of 2-stage VCO