Flip-Flop Timing, Hold Time & Latch Circuits | Lecture 20: Sequential Logic - Part 2
Overview
Sequential logic timing requires both setup and hold constraints: the minimum-delay path from one flip-flop must not deliver new data to the next flip-flop before its hold interval ends, with clock skew included. The lecture then examines transmission-gate and cross-coupled latch circuits, clock loading and transistor sizing, dynamic storage and race conditions, and clock-overlap-resistant C²MOS designs, including a dual-edge-triggered option.
Key takeaways
- Hold-time analysis is a minimum-delay problem: the contamination delay of the launching flip-flop plus the combinational path must cover the receiving flip-flop's hold time after accounting for relative clock skew.
- Clock distribution is part of system timing, not just wiring: spines and trees manage clock reach, while EDA tools handle setup and hold closure across designs too large for manual timing checks.
- A cross-coupled inverter pair stores a latch state, and combining opposite-phase latches creates an edge-triggered flip-flop; restructuring these stages can reduce clock loads substantially across millions of cells.
- Dynamic latch designs can save an inverter by relying on node capacitance, but leakage gradually erases stored charge and makes retention a design concern.
- Clock overlap can open a race path in transmission-gate flip-flops; C²MOS uses clock-controlled inverter stages to prevent direct data-to-output transfer during overlap, accepting extra devices and possible short-circuit current.
- Dual-edge-triggered flip-flops can use both rising and falling clock edges for more computation per cycle, but increase hardware and retain separate setup and hold constraints for each branch.
Chapters
- Setup time requires data to be stable before a flip-flop's clock edge; hold time requires stability after that edge.
- Clock skew means clock edges at flip-flops A and B may not align, and skew can be positive or negative depending on their relative arrival times.
- A hold violation occurs when data changes during flip-flop B's required post-clock stability interval.
- Contamination delay, written T_cd, is the minimum delay through a flip-flop or combinational logic block.
- A NAND gate's delay varies with its input transition and transistor paths, so one delay number does not describe every input case.
- Hold analysis uses minimum delay because the fastest arriving data creates the risk of changing the receiving input too soon.
- The earliest new data reaches flip-flop B after the contamination delay of flip-flop A plus the combinational block's contamination delay.
- The resulting hold constraint requires T_cd,FF_A + T_cd,com to be at least the receiving flip-flop's hold time, adjusted for the relative clock skew.
- A slower combinational path does not worsen hold timing; if the minimum-delay case is safe, longer path delays arrive later.
- A later clock at flip-flop B can extend the interval during which its input must remain stable, making hold timing harder to satisfy.
- Clock spines distribute a main clock along balanced routes; clock trees branch hierarchically to reach many flip-flops.
- Electronic design automation tools evaluate placement and identify setup and hold violations across millions of gates, a scale impractical to manage manually.
- A transmission gate uses complementary control signals to turn its NMOS and PMOS transistors on or off together.
- When the data path is enabled, two inverters pass the input to the output; when it is disabled, feedback lets the latch retain its value.
- Combining positive and negative level-sensitive latches produces an edge-triggered flip-flop, with the latch order determining the active clock edge.
- A straightforward two-latch flip-flop can connect the clock and its complement to eight transistor gates per flip-flop.
- Because transistor gates act as capacitive loads, one million such flip-flops could place roughly eight million gate loads on clock signals.
- Restructuring the latch around a cross-coupled inverter pair reduces clock-controlled gates from four to two per latch, reducing clock load across a large design.
- Connecting latch stages creates a possible reverse-conduction path from a later stage back toward the data input.
- A weaker feedback inverter I4 and stronger inverter I1 help prevent the feedback path from disturbing stored data.
- Transistor sizing can address this risk without adding a buffer, avoiding the extra devices and delay a buffer would introduce.
- Two cross-coupled NOR gates retain state when both set and reset inputs are zero.
- Set drives Q to one, while reset drives Q to zero; the stored output provides the memory behavior.
- Cross-coupling feeds a gate's output back into another gate's input, making the circuit sequential rather than purely combinational.
- A dynamic transmission-gate design removes an inverter and stores information on node capacitance instead.
- Node capacitance can be engineered through device choices such as adjusting transistor underlap, which changes source and drain capacitance.
- Stored charge leaks over time, so dynamic circuits must contend with degraded state and may require refreshing.
- An inverter delay prevents clock and clock-bar signals from switching at exactly the same instant, creating intervals where both are zero or both are one.
- Overlap can turn on both transmission gates at once, allowing input data to race through to the final output.
- One proposed workaround makes the receiving hold interval longer than the overlap period, but the lecture describes this as a fragile approach compared with preventing overlap.
- Two-phase designs use shifted clock phases rather than relying on a clock and its exact complement.
- During one phase, one latch captures data while the other retains its state; the roles reverse during the next phase.
- Separate clock-generation paths can produce the two phases, but their design and optimization add implementation complexity.
- C²MOS adds clock-controlled NMOS and PMOS devices to inverter stages, so the stages act as inverters only under the appropriate clock condition.
- The clock controls which stage is active and enables data transfer through a two-stage structure without requiring a direct data-to-output path in every clock state.
- The design adds transistors compared with simpler latch structures, trading area for improved tolerance to clock skew and overlap.
- When clock and clock-bar overlap at zero, the enabled PMOS devices may create a short-circuit current, but the circuit still blocks a direct path from data to output.
- When the signals overlap at one, the circuit likewise avoids directly connecting the input to the output.
- The brief overlap may cost power, but preventing a data race protects pipeline state and bit sequencing.
- A conventional edge-triggered flip-flop uses either rising or falling clock edges, leaving the opposite edges unused.
- A dual-edge design places two branches in parallel, with their clock controls reversed so one responds to rising edges and the other to falling edges.
- Capturing on both edges can increase throughput, but approximately doubles the branch hardware and each branch must still satisfy setup and hold timing.
Transcript excerpt
0:03 Okay. So, we have been discussing sequential logic and we actually came a long way. We discussed two major constraints, setup time constraint and hold time constraint and we basically finished discussing the setup time constraints considering a pipeline architecture where you had a flip-flop driving another flip-flop through a combinational logic block, right? We saw, basically, the constraint itself what has to happen to make things work and we saw that there is, a you, know minimum time period or maximum frequency that a system can operate, given some delay in the flip-flops and delay in the combinational logic, which is something that we have to keep in mind during design. We also talked about clock skew, basically the practical
0:50 aspect of clock distribution. So, the clock at two flip-flops will never going to be perfectly aligned with each other and we updated the constraint based on the clock skew and I told you the clock skew could be positive or negative. Don't always assume that flip-flop A will always have the clock fast and flip-flop B will always have the clock later because it is not a linear design. It is a two-dimensional complex wave-like design and it's very difficult to keep track of and it's not going to be just flip-flop A and B. It will be like thousands of flip-flop, millions of flip-flops, right? So, it's difficult or impossible to keep track of you know who comes first and who comes next. So, basically skew is a parameter that could be positive or negative with respect to one another. So, you just have to update the sign of the skew accordingly. Okay. So, today we'll start with the
1:38 whole time constraint, the other end of the constraint and and go from there. I hope you remember by now even without looking at
This is the start of the transcript. Save this video free for the whole transcript, searchable and time-coded.
In this course
Related notes
Summary, takeaways, and chapters were generated by AI from the video's transcript and may contain errors. The video belongs to its creator, NorDIC Lab.