Area overhead

Each scan flop adds one 2-to-1 mux on its data input. In the course teaching library a DFF costs 8 area units and a MUX2 costs 5, so a scan flop costs 13 units.

The calc block prices one thousand flops. The muxes add 5000 library units, a 62.5% overhead on the flop array before routing.

course.lib teaching areas (library units, not a process): DFF = 8, MUX2 = 5
scan flop = 8 + 5 = 13 units
1000 plain flops = 1000 x 8 = 8000 units
1000 scan flops = 1000 x 13 = 13000 units
mux overhead = 13000 - 8000 = 5000 units = 5000 / 8000 = 62.5%

Timing overhead

The scan mux adds delay to the functional data path. Arrival grows by the mux delay and slack, defined as required minus arrival, shrinks by the same amount.

The calc block uses stated EXAMPLE parameters. SE itself switches only between phases, so it carries no cycle-to-cycle timing pressure.

stated EXAMPLE parameters: scan mux delay 0.04 ns, required time 1.50 ns
arrival before scan = 1.20 ns
slack before = 1.50 - 1.20 = 0.30 ns
arrival after scan mux = 1.20 + 0.04 = 1.24 ns
slack after = 1.50 - 1.24 = 0.26 ns
slack lost to the mux = 0.30 - 0.26 = 0.04 ns

Test time and routing cost

Test time is shift cycles times the shift-clock period, usually a slow EXAMPLE tester rate. Extra chains cut cycles but each chain needs pins and routing across the floorplan.

The calc block extends the [4, 3, 3] example to five patterns. Forty-five cycles at 40 ns need 1.8 us of tester time.

L = 4, per pattern = 4 + 1 + 4 = 9 cycles
5 patterns = 5 x 9 = 45 cycles
EXAMPLE shift period 40 ns (stated example parameter)
test time = 45 x 40 ns = 1800 ns = 1.8 us

Related topics

• Chain Stitching and Balancing

• Area, Power and Timing Trade-offs