G9.16.1 A Read That Does Not Flip the Cell

HardImplementngspice 42 + SKY130 · serverReview the topic

Size a 6T SRAM cell so that a read barely disturbs it and is fast, while a write still flips it.

Circuit context: The bit cell and sense amplifier of a small SRAM macro. Size an SRAM cell for read stability and write ability.

A small SRAM macro must keep a large read margin at a low supply. The first cell layout uses equal widths, and its reads disturb the stored value far too much.

This is an Implement problem. Size the six transistors of the block cell. The check reads a stored 0 and a stored 1, writes both values, and measures the gate area.

In either read the 0 node may rise to 0.12 V at most. The bit line must fall 100 mV within 42 ps. Each write must flip the cell while its bit line is still at 0.45 V or more. The gate area is at most 0.75 µm².

What the Grader Measures

Your subcircuit cell (ports q, qb, bl, blb, wl, vdd) under the grader's own bench, one run per corner (tt):

CheckCornerRule
Highest 0-node voltage in either readtt≤ 120 mV
Slower bit line to fall 100 mV after the word linett≤ 42 ps
Lower bit-line voltage at which a write flips the celltt≥ 450 mV
Budget: Gate areatt≤ 750 mµm², the lower the better
Full specification
The block that is checked (keep the name and the ports)
.subckt cell q qb bl blb wl vdd
* two cross-coupled inverters and two access NMOS
.ends
CheckRuleHow it is measured
0-node voltage in a read≤ 0.12 Vthe higher peak of two reads (a stored 0 and a stored 1). Bit lines: 50 fF at 1.8 V
Read speed≤ 42 psfrom the word line at 0.9 V to the discharging bit line at 1.7 V; the slower of the two reads
Writebit line ≥ 0.45 V at the flipone bit line ramps from 1.8 V to 0 in 20 ns. Its voltage when q and qb cross; the lower of two writes
Gate area (budget)≤ 0.75 µm²the sum of W · L over the transistors of cell

The check runs once at tt, 27 °C and 1.8 V.

Examples

Example 1

The Starter: Every Width 0.5 µm
Design
.subckt cell q qb bl blb wl vdd
XU1 q qb vdd vdd sky130_fd_pr__pfet_01v8 W=0.5 L=0.15
XD1 q qb 0 0 sky130_fd_pr__nfet_01v8 W=0.5 L=0.15
XA1 bl wl q 0 sky130_fd_pr__nfet_01v8 W=0.5 L=0.15
XU2 qb q vdd vdd sky130_fd_pr__pfet_01v8 W=0.5 L=0.15
XD2 qb q 0 0 sky130_fd_pr__nfet_01v8 W=0.5 L=0.15
XA2 blb wl qb 0 sky130_fd_pr__nfet_01v8 W=0.5 L=0.15
.ends
Measured by the grader
Highest 0-node voltage in either read: 309.8 mV (needs ≤ 120 mV)
Slower bit line to fall 100 mV after the word line: 50.45 ps (needs ≤ 42 ps)
Lower bit-line voltage at which a write flips the cell: 684.9 mV (needs ≥ 450 mV)
Gate area: 450 mµm² (budget ≤ 750 mµm²)

The 0 node rises to 0.310 V in a read: far too close to the other inverter's threshold.

Example 2

Access 1 µm, Pull-Down 0.42 µm
Design
.subckt cell q qb bl blb wl vdd
XU1 q qb vdd vdd sky130_fd_pr__pfet_01v8 W=0.42 L=0.15
XD1 q qb 0 0 sky130_fd_pr__nfet_01v8 W=0.42 L=0.15
XA1 bl wl q 0 sky130_fd_pr__nfet_01v8 W=1 L=0.15
XU2 qb q vdd vdd sky130_fd_pr__pfet_01v8 W=0.42 L=0.15
XD2 qb q 0 0 sky130_fd_pr__nfet_01v8 W=0.42 L=0.15
XA2 blb wl qb 0 sky130_fd_pr__nfet_01v8 W=1 L=0.15
.ends
Measured by the grader
Highest 0-node voltage in either read: 531.8 mV (needs ≤ 120 mV)
Slower bit line to fall 100 mV after the word line: 41.88 ps (needs ≤ 42 ps)
Lower bit-line voltage at which a write flips the cell: 847.2 mV (needs ≥ 450 mV)
Gate area: 552 mµm² (budget ≤ 750 mµm²)

A fast read, but the 0 node reaches 0.532 V.

Example 3

Pull-Ups of 1 µm
Design
.subckt cell q qb bl blb wl vdd
XU1 q qb vdd vdd sky130_fd_pr__pfet_01v8 W=1 L=0.15
XD1 q qb 0 0 sky130_fd_pr__nfet_01v8 W=1.5 L=0.15
XA1 bl wl q 0 sky130_fd_pr__nfet_01v8 W=0.42 L=0.15
XU2 qb q vdd vdd sky130_fd_pr__pfet_01v8 W=1 L=0.15
XD2 qb q 0 0 sky130_fd_pr__nfet_01v8 W=1.5 L=0.15
XA2 blb wl qb 0 sky130_fd_pr__nfet_01v8 W=0.42 L=0.15
.ends
Measured by the grader
Highest 0-node voltage in either read: 89.5 mV (needs ≤ 120 mV)
Slower bit line to fall 100 mV after the word line: 44.04 ps (needs ≤ 42 ps)
Lower bit-line voltage at which a write flips the cell: 289.8 mV (needs ≥ 450 mV)
Gate area: 876 mµm² (budget ≤ 750 mµm²)

With 1 µm pull-ups the reads are safe. But a write only flips the cell when its bit line is down to 0.290 V.

Constraints and Contract

  • The netlist defines .subckt cell q qb bl blb wl vdd ... .ends exactly once, with these ports in this order.
  • Inside cell: X instances of sky130_fd_pr__nfet_01v8 or sky130_fd_pr__pfet_01v8, with comments and .param lines. No sources, resistors or capacitors. Use m= for parallel copies; mult= is refused.
  • The check writes the .lib line itself and sets the starting state of q and qb.

Hints

Hint 1

In a read, the access transistor and the pull-down form a divider on the 0 node. Which one must be stronger, and by how much for 0.12 V?

Hint 2

In a write, the access transistor fights the pull-up. Keep the pull-up the weakest of the three.

Hint 3

Both halves are read and written. A cell that is strong on one side only fails on the other.

What the grader checks: The check places four copies of cell under its own bench at tt and 27 °C. Two are reads with 50 fF bit lines and two are writes with a slow ramp, one for each stored value. The tool is ngspice 42 (libngspice0 42+ds-3build1) with the pinned SKY130 models, through tools/spice-judge.mjs. Mismatch between the halves, corners and real array bit lines are outside this problem.

Run simulates your netlist as written, with your own analyses and measurements. Submit sends it to the grader, which keeps only the subcircuit cell and makes 1 ngspice run. The numbers come from ngspice and the SKY130 models the site pins; they prove the measured cases, not every condition a chip meets.

See all 1 steps and the final capstone

Report a problem with this page

Your netlist SPICESaved locally
Ln 1, Col 1Tab indents · Ctrl/⌘ + Enter runs
Not run yet
Run your netlist to see its measurements and the ngspice log here.
ngspice 42 + SKY130 · server