Skip to main content

The SKY130 passive catalogue

This page lists the resistors and capacitors SKY130 gives you, what each one costs in area, and where each number comes from. It’s meant to stay open beside your tools while you design.

We measured every resistance and capacitance below in hpretl/iic-osic-tools:2026.08 with ngspice 47: we instantiated the device at a stated geometry, simulated it and divided, rather than copying the value from a datasheet. Every parameter is quoted along with the file it lives in, so you can check any line on this page with one grep.

Units of W and L

W and L on a SKY130 device are plain numbers in microns, with no unit suffix. W=1 is one micrometre, and W=1u asks for a millionth of that. On a resistor the run doesn’t stop: a strip written W=1u L=10u comes out at 3193.812 Ω where W=1 L=10 gives 3550.443 Ω, with no error, and ngspice exits with status 0. The same rule applies to a capacitor’s W and L and to a diode’s area and perim. Getting started shows what happens on a resistor, and AD103’s Reading a SKY130 device model has the exact error text for transistors.

Resistors

Resistor table

The table has thirteen devices in nine rows, and their sheet resistances span five orders of magnitude. R measured is the resistance of one square (a strip drawn W = 1 µm, L = 1 µm), measured at 0.1 V:

Device What it is physically \(R_\square\) from the PDK R measured, W=1 L=1
res_xhigh_po poly with the ultra-high-resistance implant 2000 Ω/□ 2152.022 Ω
res_high_po poly with the high-resistance implant 317.3885 Ω/□ 695.4646 Ω
res_generic_pd p+ diffusion 197 Ω/□ 195.4779 Ω
res_generic_nd n+ diffusion 120 Ω/□ 117.4914 Ω
res_generic_po ordinary silicided poly, the same material as the gate 48.2 Ω/□ 50.92314 Ω
res_generic_nw the n-well itself 1700 Ω/□ 1692.556 Ω
res_iso_pw isolated p-well 3816 Ω/□ 14969.38 Ω at W=2.65 L=10
res_generic_l1 local interconnect 12.2 Ω/□ 11.97427 Ω
res_generic_m1 … m5 metal 1–5 0.125 / 0.125 / 0.047 / 0.047 / 0.0285 Ω/□ 0.1288327 Ω (m1), 0.04770963 Ω (m3)

The first five devices and res_iso_pw are subcircuits, so you instantiate them on X lines with W= and L= in capitals. res_generic_nw, res_generic_l1 and the metal resistors are bare .model cards, so they go on R lines with lower-case w= and l=:

XR1 a 0 0 sky130_fd_pr__res_high_po W=1 L=10 mult=1
r1  b 0   sky130_fd_pr__res_generic_m1 w=1 l=1

If you get that backwards, ngspice says Error: unknown subckt: or warning, can't find model. The sheet resistances in the table come from these files in the image:

/foss/pdks/sky130A/libs.tech/ngspice/r+c/res_typical__cap_typical.spice
        rp1=48.2  rdn=120  rdp=197  rnw=1700  rl1=12.2
        rm1=0.125  rm2=0.125  rm3=0.047  rm4=0.047  rm5=0.0285
        rspwres=3816
/foss/pdks/sky130A/libs.ref/sky130_fd_pr/spice/sky130_fd_pr__res_high_po.model.spice
        rsheet = 317.3885     rhead_ps = 345.8312
/foss/pdks/sky130A/libs.ref/sky130_fd_pr/spice/sky130_fd_pr__res_xhigh_po.model.spice
        rsheet = 2000.0

Magic’s extraction technology file, /foss/pdks/sky130A/libs.tech/magic/sky130A.tech, keeps its own copy of these values in milliohms per square (xhrpoly 319800, uhrpoly 2000000, mrp1 48200), and it doesn’t agree with the ngspice file everywhere. Ohms per square explains which one to use when.

End resistance of res_high_po

A real resistor has to be contacted, so the device is a strip of resistive poly with a wide head at each end that holds a row of contacts, and those heads add resistance of their own. If you increase the length by 10 µm, the difference tells you what the body alone costs:

Device R at L = 1 R at L = 11 slope (Ω per µm of L)
res_high_po, W=1 695.4646 Ω 3867.662 Ω 317.2198
res_xhigh_po, W=1 2152.022 Ω — —

That’s 317.2198 Ω per micron of length at W = 1 µm, against rsheet = 317.3885 on the model card. The two are 0.05 % apart, so the body really is the sheet resistance the card promises. The remaining \(695.4646 - 317.2198 = 378.24\ \Omega\) is end resistance: the contact head (299.8915 Ω, see Ohms per square) plus a short body extension that the model adds. The end resistance stays the same when you shorten the resistor, so in a one-square res_high_po the ends contribute more than the body does.

The design rule that follows is never draw a short precision resistor. The ends are 378 Ω whatever you do, so they make up 54 % of the value at one square, 19 % at five squares and 11 % at ten squares. Below about five squares, a large fraction of your resistor is contact resistance, and contact resistance matches far worse than sheet resistance does.

The fixed-width precision family

res_high_po and res_xhigh_po also ship in five pre-characterised widths each. These are the versions a layout tool will give you, because their contact head geometry is fixed:

Width (µm) res_high_po_* Ω/µm of length res_xhigh_po_* Ω/µm of length
0.35 1085.483 5713.715
0.69 472.111 2898.261
1.41 219.662 1418.298
2.85 105.534 701.684
5.73 56.460 349.005

Each figure is (R at L=11 − R at L=1) / 10 from a measured pair of resistors, so the contact heads have already been subtracted out.

If you multiply each figure in the xhigh column by its width, you get 1999.80 Ω/□ for all five (\(0.35 \times 5713.715 = 1999.80\), \(5.73 \times 349.005 = 1999.80\)), against rsheet = 2000.0 in the model file. All five widths are the same material described by the same sheet resistance.

The same calculation on the high column gives 379.9, 325.8, 309.7, 300.8 and 323.5 Ω/□, which don’t agree with each other. The narrow-width flavours are separately fitted devices, and res_high_po_0p35 behaves like a 380 Ω/□ material rather than a 317 Ω/□ one. So if you move a design from one width to another, re-simulate it instead of scaling the old value by hand.

10 kΩ four ways

Here is the same 10 kΩ target in four flavours, with each length tuned until the simulator agreed:

Device W (µm) L (µm) R measured area
res_xhigh_po_0p35 0.35 1.647 9998.007 Ω 0.576 µm²
res_high_po_0p35 0.35 8.351 10000.90 Ω 2.923 µm²
res_generic_nd 1 85.1 9998.521 Ω 85.1 µm²
res_generic_po 1 196.4 10001.30 Ω 196.4 µm²

That’s a factor of 341 in area for the same 10 kΩ, and it’s the reason a process bothers to have a high-resistance implant at all: without one, a modest bias resistor is bigger than a logic gate.

For scale, here is the smallest inverter’s size, from /foss/pdks/sky130A/libs.ref/sky130_fd_sc_hd/lef/sky130_fd_sc_hd.lef:

MACRO sky130_fd_sc_hd__inv_1
  SIZE 1.380 BY 2.720 ;

That’s 3.7536 µm² for the smallest inverter in the standard-cell library, so the 10 kΩ res_xhigh_po_0p35 is smaller than one inverter, while the res_generic_po version takes the area of fifty-two of them. Scale that up to 100 kΩ in ordinary poly (1964 µm², about 520 inverters) and the resistor is bigger than the circuit it biases, so choosing the resistor flavour is a design decision in its own right.

Capacitors

Capacitor table

Each capacitor was driven by a 1 V AC source at 1 kHz, and its value computed as \(C = |I| / (2\pi f V)\) with \(V = 1\) V:

Device Geometry C measured density
cap_mim_m3_1 2 × 2 µm 9.302250 fF 2.326 fF/µm²
cap_mim_m3_1 10 × 10 µm 206.5822 fF 2.066 fF/µm²
cap_mim_m3_1 22.2 × 22.2 µm 1000.314 fF 2.030 fF/µm²
cap_mim_m3_2 10 × 10 µm 206.5822 fF 2.066 fF/µm²
cap_vpp_11p5x11p7_l1m1m2m3m4_shieldm5 11.5 × 11.7 µm (fixed) 147.3400 fF 1.095 fF/µm²
nfet_01v8 as a MOS cap, gate at 1.8 V 10 × 10 µm gate 787.6883 fF 7.877 fF/µm²
the same device, gate at 0 V 10 × 10 µm gate 220.7182 fF 2.207 fF/µm²
cap_var_lvt at 1.8 V 10 × 10 µm 817.8665 fF 8.179 fF/µm²
the same varactor at 0 V 10 × 10 µm 99.51349 fF 0.995 fF/µm²

A few things in the table are worth pointing out:

  1. m3_1 and m3_2 are the same capacitor built on different metals. cap_mim_m3_1 sits between met3 and the capm plate, and cap_mim_m3_2 sits between met4 and cap2m. Their capacitance is identical to seven digits, so you choose between them by what you need to route underneath.
  2. MIM density rises as the plate shrinks, because the perimeter term is fixed per micron of edge and a smaller plate has more edge for its area: 2.326 fF/µm² at 2 × 2 µm against 2.030 at 22.2 × 22.2 µm.
  3. A MOS capacitor is 3.8× denser than a MIM, but its value depends on the voltage across it: 787.6883 fF at 1.8 V and 220.7182 fF at 0 V, a factor of 3.57 apart on the same physical device. Use it for supply decoupling, where one plate is always at the rail, and never in a filter whose corner frequency you care about.
  4. The varactor uses that nonlinearity on purpose. It goes from 99.5 fF to 817.9 fF, a factor of 8.22, over the supply range, which makes it the tuning element in an oscillator.

MIM capacitance from the model parameters

The MIM model comes down to three lines of /foss/pdks/sky130A/libs.ref/sky130_fd_pr/spice/sky130_fd_pr__cap_mim_m3_1.model.spice:

.param wc = 'w+m3_dw*1e6+tol_m3*1e6'
.param carea  = 'camimc*(wc)*(lc)'
.param cperim = 'cpmimc*((wc)+(lc))*2'

(lc is defined the same way as wc, on the line immediately after it.)

and three constants, which live in /foss/pdks/sky130A/libs.tech/ngspice/r+c/res_typical__cap_typical__lin.spice and .../sky130_fd_pr__model__r+c.model.spice:

+ camimc=  2.00e-15  ; Units: farad/micrometer^2
+ cpmimc = 0.19e-15 ; Units: farad/micrometer
+ m3_dw = -0.025u

So when you draw a plate 10 µm square, the metal etches back by 0.025 µm and the real plate is 9.975 µm on a side:

\[2.00 \times 9.975^2 \;+\; 0.19 \times 2 \times (9.975 + 9.975) = 199.00125 + 7.581 = \boxed{206.582\ \text{fF}}\]

ngspice measured 2.065822e-13, which matches to six significant figures using only three constants and a subtraction. When a PDK number and your own arithmetic agree that closely, you can be confident you understand how the device is modelled.

What 1 pF costs

Device Area for 1 pF Caveat
MOS cap at 1.8 V 127 µm² only if one plate stays at a rail
MIM 493 µm²: 22.2 × 22.2 µm, measured at 1000.314 fF linear; the usual choice for a signal path
VPP fringe 913 µm² it uses met3 and met4, the MIM’s own layers, so it can’t share area with a MIM

493 µm² is about the area of 130 inverters. It’s the number to bring to any “just put a 1 nF cap on it” conversation, because a nanofarad of MIM is half a square millimetre, the area of about 130 000 inverters. What a picofarad costs goes through this at length.

Parasitic capacitance to substrate

Every layer has capacitance to the substrate, whether you asked for it or not. These values come from the defaultareacap lines in Magic’s /foss/pdks/sky130A/libs.tech/magic/sky130A.tech, which gives them in aF/µm²:

Layer to substrate aF/µm² as fF/µm²
poly 106.13 0.106
local interconnect (li) 36.99 0.037
met1 25.78 0.026
met2 17.5 0.018
met3 12.37 0.012
met4 8.42 0.008
met5 6.32 0.006
n-well / deep n-well 120 0.120

So a 100 µm × 1 µm met1 wire carries 2.6 fF to ground, which is more than a quarter of a 10 fF capacitor you might draw on purpose elsewhere. Parasitic capacitance covers this in more detail, including the sidewall terms that this table leaves out.

Inductors

SKY130 has no good inductor. It ships three spiral models (sky130_fd_pr__ind_03_90, ind_05_125 and ind_05_220). ind_05_220 is 9.92 nH, occupies tens of thousands of µm², and has Q = 0.0000150516 at 1 kHz. The best Q any of the three ever reaches is 21.1, at 6.38 GHz, and above 3.52777 GHz the reactance of ind_05_220 changes sign. Inductors on a chip and What to build instead of an inductor cover why, and what replaces it.

Corners

tt is the typical corner, for transistors and passives alike. Passives also have corners of their own, selected by the same word at the end of the .lib line: ll (low resistance, low capacitance) and hh (high resistance, high capacitance), plus hl and lh.

We measured these spreads rather than reading them off a parameter file, by simulating the same four devices three times and changing only the corner name:

Device ll tt hh spread about tt
res_high_po W=1 L=10 3106.637 Ω 3550.443 Ω 3994.248 Ω ±12.5 %
res_xhigh_po W=1 L=10 18037.95 Ω 21221.12 Ω 24404.29 Ω ±15.0 %
res_generic_po W=1 L=10 451.1426 Ω 509.2314 Ω 583.1667 Ω −11.4 % / +14.5 %
cap_mim_m3_1 10 × 10 179.7322 fF 206.5822 fF 233.8667 fF −13.0 % / +13.2 %

The parameters behind those runs are set in /foss/pdks/sky130A/libs.tech/ngspice/r+c/res_low__cap_low.spice, res_typical__cap_typical.spice, res_high__cap_high.spice and their __lin siblings:

Parameter ll tt hh
rp1 (ordinary poly) 44 48.2 53.52
rdn (n+ diffusion) 111.6 120 128.4
rdp (p+ diffusion) 175.3 197 218.7
rl1 (local interconnect) 10.31 12.2 14.02
rm1 (metal 1) 0.111 0.125 0.139
rnw (n-well) 1378 1700 2022
sky130_fd_pr__res_high_po__var −0.125 0.0 +0.125
camimc (MIM) 1.778e-15 2.00e-15 2.231e-15

A device doesn’t always move by exactly its parameter’s percentage. rp1 shifts by −8.7 % and +11.0 %, but a real res_generic_po moves by −11.4 % and +14.5 %, because the corner also changes the etch bias that sets the strip’s effective width, so simulate the corner rather than scaling the typical answer by hand.

Every device measured above moves by more than ±10 % between corners. A ratio of two of the same device stays put, which is why analog design is built on ratios; see Matching and Pelgrom’s law.

Re-running these numbers

The numbers come from two sets of decks in labs/passives-decks:

cd labs/passives-decks
make            # the guide pages' decks: head/body, corners, MIM, spirals, mismatch
make catalogue  # this page's extra devices: the seven materials at one square,
                # both fixed-width families, 10 kΩ four ways, and the MIM/varactor rows

make compares its results against 45 stored values and ends with PASS or FAIL. make catalogue prints the numbers from spice/catalogue.spice for you to read against the tables above, and it takes about a minute.

You don’t need any environment setup, because every deck names the model library by its absolute path, so a bare container works. The related labs are Lab 01: A resistor you designed, Lab 02: How big is a picofarad? and Lab 04: Two resistors that disagree.