Pick one unit or two units behind a multiplexer, and pick the supply voltage. A regulator floor limits the voltage, and the least energy per operation wins.
Circuit context: Inverter chain standing in for a critical path on a scalable supply. Choose the unit count and supply that meet a throughput at the least energy.
A datapath must finish one operation every 3.5 ns. Its critical path is a chain of 24 inverters, and the chip may come from the slow corner. Lesson W2.7 built the same two architectures around a shorter chain.
Architecture single is one unit that takes every operation. Architecture dual is two units side by side, each taking every second operation. A 2:1 multiplexer picks the result that is due.
The block shares a regulator that cannot go below 1.20 V, because other circuits on the rail need it. Choose the architecture and the voltage. Every result must arrive in time on a slow chip, and a typical chip must spend at most 180 fJ per operation.
What the Grader Measures
Your subcircuit rail (ports single, dual) under the grader's own bench, one run per corner (ss, tt):
| Check | Corner | Rule |
|---|---|---|
| Wrong or late results on a slow chip | ss | ≤ 0 |
| Supply voltage | tt | ≥ 1.2 V |
| Budget: Energy per operation, typical chip | ss | ≤ 180 fJ, the lower the better |
Full specification
.subckt rail single dual
V1 <single or dual> 0 <your voltage>
.ends
| Check | Rule | How the grader measures it |
|---|---|---|
| Results on time, slow chip | 0 wrong results | ss corner, 6 operations. Each result is read 50 ps before its deadline and must sit on the right side of half the supply. |
| Supply inside the regulator range | ≥ 1.20 V | The voltage on the node you power |
| Energy per operation, typical chip (budget) | ≤ 180 fJ | tt corner: the supply voltage times the current of V1, integrated over 4 operations and divided by 4 |
Both architectures are built by the grader; only the one your source powers runs. The multiplexer, its select driver and both units draw from your source. The inputs come from ideal sources and are not counted.
The grader runs at 27 °C. Each unit is 24 inverters (W_N = 1 µm, W_P = 2 µm, 5 fF on each stage), as in lesson W2.7.
Examples
Example 1
One Unit at the Nominal 1.8 V- Design
.subckt rail single dual V1 single 0 1.8 .ends
- Measured by the grader
Wrong or late results on a slow chip at ss: 0 (needs ≤ 0) Supply voltage at tt: 1.8 V (needs ≥ 1.2 V) Energy per operation, typical chip: 404.9 fJ (budget ≤ 180 fJ)
Fast on every chip, and more than twice the energy budget.
Example 2
Two Units at 1.8 V- Design
.subckt rail single dual V1 dual 0 1.8 .ends
- Measured by the grader
Wrong or late results on a slow chip at ss: 0 (needs ≤ 0) Supply voltage at tt: 1.8 V (needs ≥ 1.2 V) Energy per operation, typical chip: 445.5 fJ (budget ≤ 180 fJ)
Two units only save energy when they let the voltage fall.
Example 3
Two Units at 1.05 V- Design
.subckt rail single dual V1 dual 0 1.05 .ends
- Measured by the grader
Wrong or late results on a slow chip at ss: 4 (needs ≤ 0) Supply voltage at tt: 1.05 V (needs ≥ 1.2 V) Energy per operation, typical chip: 150.3 fJ (budget ≤ 180 fJ)
Cheap, below the regulator range, and too slow on a slow chip.
Constraints and Contract
- The netlist defines .subckt rail single dual ... .ends exactly once, with these ports in this order.
- Inside rail: one DC voltage source named V1, from the port of your architecture to ground (node 0). Comments and .param lines are allowed. A pwl, pulse or sin source is refused.
- The grader writes the .lib line itself; everything outside rail is replaced by its bench.
Hints
Hint 1
Start where the lesson ended: try dual, and find its lowest voltage on a slow chip. Then compare it with the regulator range.
Hint 2
If both architectures end up at the same voltage, what does the second unit buy? Count what the multiplexer and its select line add.
Hint 3
Check timing at ss and energy at tt for each candidate. One unit may meet 3.5 ns at the floor; find out, do not assume.
What the grader checks: The grader places your rail under its own bench, once at ss and once at tt, at 27 °C. It runs on ngspice 42 (libngspice0 42+ds-3build1) with the pinned SKY130 models. The unit is an inverter chain standing in for a critical path. Wires, input registers and the clock of the operands are not modelled.
Run simulates your netlist as written, with your own analyses and measurements. Submit sends it to the grader, which keeps only the subcircuit rail and makes 2 ngspice runs. 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