Skip to main content

Lab 01: Open a cell you did not draw

You’ll load this course into the workbench, open the foundry’s own inverter, sky130_fd_sc_hd__inv_1, in Magic, and measure its two transistors from the layout file. Then you’ll run four checks on the same cell (DRC under two rule sets, extraction, and LVS) and get them to agree about it. If they do, you know your toolchain works, so any problem you run into after this lab is a problem in your drawing.

Package: labs/lab-01-open-the-cell/

Before you start

Time: about 1 hour 15 minutes, plus 20 minutes for the optional SiliWiz part. make itself takes about a second and a half.

Part 1: Load the course

If you haven’t set up the workbench yet, do IC101 Lab 03 first.

  1. Start the noVNC desktop the way IC101 showed you, and open a terminal on it. Type every command in this lab in that terminal, not on your laptop.

  2. Load the course environment and check the PDK. The setting lasts only as long as the terminal you load it in, so you’ll need to load it again in every new terminal you open.

    . /foss/designs/common/.designinit
    echo $PDK

    echo $PDK must print sky130A. If it prints ihp-sg13g2, which is the image’s default, the first line didn’t run in this terminal.

  3. Get the course files with mod, a command that .designinit defines.

    mod add ad104     # first time only
    mod ad104         # puts you in the course folder
  4. Check the versions of the three tools whose output this course prints:

    magic --version
    netgen -batch quit | grep '^Netgen'
    ngspice --version | grep ngspice-
    8.3.681
    Netgen 1.5.323 compiled on Tue Aug 25 11:33:45 AM CEST 2026
    ** ngspice-47 : Circuit level simulation program

    If you have a different version, your outputs may not match the ones printed in these labs.

What you see What it means
.designinit: No such file or directory The workspace isn’t set up, so do IC101 Lab 03. Until then you can still run make in any lab, because every lab package pins its own PDK.
couldn't read Username for 'https://github.com' on mod add ad104 The repo didn’t answer an anonymous request. That’s a problem on our side rather than anything you did, so please tell us in Discord.

For anything else, see IC101’s Troubleshooting page.

You should see sky130A from step 2 and the three versions from step 4.

Part 2: Look at the cell in Magic

What is in the foundry’s inverter, and what is missing from it?

  1. From the course folder, go into the lab package and open the cell:

    cd labs/lab-01-open-the-cell
    make gui

    (make edit does the same thing, and it’s the name every other lab uses.) Magic opens two windows with the cell loaded: layout1, which shows the drawing, and tkcon, a Tcl console where you type Magic commands. You get the terminal back while Magic stays open, so you can keep using both.

  2. Using Anatomy of a standard cell as a map, find the items below. The figure in What a layout is shows which colour is which layer. To check a position or a size, put the cursor box on it (left-click its lower-left corner, right-click its upper-right) and type box in the console, which prints the box’s width, height and corners in microns.

    • the n-well across the top, and the p-diffusion inside it
    • the n-diffusion in the bottom half
    • the one poly stripe that crosses both diffusions, which is why the two gates are connected without a wire
    • the two metal1 rails, straddling y = 0 and y = 2.72
    • the contacts, each of which is 0.17 µm square
  3. Now look for something the cell doesn’t have, a well tap. A tap is a contacted strip of ptap or ntap, which is silicon doped the same type as the well under it and tied to a rail. Source and drain contacts don’t count as taps, because their diffusion is the opposite type to the well. (What a layout is explains what a tap is for.)

You should see every item in step 2, and no tap anywhere in the cell. Leave Magic open, because you’ll use it again in Part 6.

Part 3: Measure the transistors

How wide and how long are the two transistors? You can work this out yourself from the layout file, before any tool extracts it for you.

  1. In the terminal, print the file’s header and its two transistor layers:

    F=$PDK_ROOT/sky130A/libs.ref/sky130_fd_sc_hd/mag/sky130_fd_sc_hd__inv_1.mag
    head -4 $F
    grep -A1 -e '<< scnmos >>' -e '<< scpmoshvt >>' $F
    magic
    tech sky130A
    magscale 1 2
    timestamp 1785844050
    << scnmos >>
    rect 120 47 150 177
    << scpmoshvt >>
    rect 120 297 150 497

    scnmos is the NMOS and scpmoshvt the high-threshold PMOS. Each rect line is one rectangle, given as x1 y1 x2 y2, and it covers the patch of diffusion under the poly stripe, which is the transistor’s channel (What a layout is).

  2. The header says magscale 1 2, so one unit in the file is 0.005 µm (Drawing in Magic). For each rectangle, work out its height and its width in µm, and write them down. The height is the transistor’s W and the width is its L.

You should see two rect lines, and four numbers in your notes: W and L for each transistor. Part 4 checks them against what the tools measure.

Part 4: Run the check

Do the tools measure the cell the way you did?

  1. Run the lab:

    make

    The Makefile sets PDK=sky130A itself, so this works even if .designinit failed for you. It takes about a second and a half (1.35, 1.46 and 1.60 s on three timed clean runs) and prints:

    == magic: load, DRC, extract   (sky130_fd_sc_hd__inv_1)
    == netgen: layout vs the foundry's own netlist
    
      THE FOUNDRY'S INVERTER, MEASURED BY YOUR TOOLS
    
      ok  cell width                              1.38 um   (reference 1.38)
      ok  cell height                             2.72 um   (reference 2.72)
      ok  DRC violations, default style                 0   (reference 0)
      ok  DRC violations, full style                    3   (reference 3)
      ok  devices extracted                             2   (reference 2)
    
          cell area  1.38 x 2.72 = 3.7536 um^2
    
      THE TWO TRANSISTORS IT FOUND
          X0  sky130_fd_pr__nfet_01v8          W=0.65  L=0.15   drain area 0.169 um^2
          X1  sky130_fd_pr__pfet_01v8_hvt      W=1  L=0.15   drain area 0.26 um^2
    
      LVS: Circuits match uniquely
    
      THE THREE THINGS THE FULL RULE SET COMPLAINS ABOUT
          All nwells must contain metal-connected N+ taps (nwell.4)
          P-diff distance to N-tap must be < 15.0um (LU.3)
          N-diff distance to P-tap must be < 15.0um (LU.2)
          ...
    
    PASS  Magic read the foundry's own layout, found zero violations under the
          default rule set and the three expected context rules under the full
          one, pulled two transistors out of the geometry, and netgen agreed
          they are the circuit the foundry says the cell is.

    If it stops or a number differs:

    What you see What it means
    DRC violations, default style is not 0 The technology did not load. Check that $PDK_ROOT points at the PDKs (/foss/pdks in the workbench image) and that .magicrc is in the folder you ran make from.
    results/report.txt is missing Magic didn’t finish, so read results/magic.log.
    LVS: netgen wrote no report Read results/netgen.log. The cause is usually $PDK_ROOT again.
    magic: command not found You’re not in the workbench container.
  2. Compare the two transistor lines with the W and L you wrote down in Part 3.

  3. Look at the count under the full rule set. The foundry’s own cell fails the full rule set, and that’s the answer we expect. (Magic calls a rule set a DRC style, which is why the output says default style and full style.) Why does the full rule set find three violations in a cell that has been taped out hundreds of times?

    All three come from the same omission. The cell has no well tap, because the library puts its taps in a separate cell, and Anatomy of a standard cell explains why it’s done that way. make prints the same explanation under the three rule names, where the output above shows ....

You should see PASS, 0 under the default rule set and 3 under the full one, and W=0.65 L=0.15 and W=1 L=0.15 matching the numbers you wrote down.

Part 5: Read the extracted netlist

What circuit did Magic pull out of the geometry?

  1. Print the netlist Magic extracted:

    make netlist
    * NGSPICE file created from sky130_fd_sc_hd__inv_1.ext - technology: sky130A
    
    .subckt sky130_fd_sc_hd__inv_1 A VGND VNB VPB VPWR Y
    X0 Y A VGND VNB sky130_fd_pr__nfet_01v8 ad=0.169 pd=1.82 as=0.169 ps=1.82 w=0.65 l=0.15
    X1 Y A VPWR VPB sky130_fd_pr__pfet_01v8_hvt ad=0.26 pd=2.52 as=0.26 ps=2.52 w=1 l=0.15
    .ends
  2. Count the ports; there are six. A and Y are the input and output, and VPWR and VGND are the supplies. What are VNB and VPB, and why are they pins?

    They’re the bulk terminals, the fourth terminal on each device line. VNB is the NMOS body (the p-substrate) and VPB the PMOS body (the n-well). The cell exposes them as pins because it doesn’t tie them to anything itself, since it has no well tap, as you found in Part 2.

  3. Find w=0.65 and w=1. Nobody typed those numbers in; Magic got them by measuring the same two rectangles you measured in Part 3.

  4. Find pfet_01v8_hvt. It tells you the PMOS is the high-threshold flavour, which you’ll need to remember in Lab 03.

You should see two devices, six ports, and w=0.65 l=0.15 on the NMOS.

Part 6: Run the flow by hand

Which commands does Magic run for make, and which of them are easy to get wrong?

Everything Magic did for make in Part 4 comes down to a short list of commands that you can type yourself. Every remaining lab in the course uses this same flow, so it’s worth running it once with your own fingers.

  1. In the tkcon console (the Tcl console window that opened with Magic in Part 2), type these one at a time:

    select top cell
    drc euclidean on
    drc check
    drc catchup
    drc count total
    drc style drc(full)
    drc check
    drc catchup
    drc count total
    drc why
    extract do local
    extract all
    ext2spice lvs
    ext2spice -o /tmp/inv.spice
  2. Watch the count change when you switch rule sets. The first drc count total prints Total DRC errors found: 0, the second prints Total DRC errors found: 3, and drc why then names the three rules.

  3. The last line wrote the extracted netlist to /tmp/inv.spice. In the terminal, compare it with the one make wrote:

    diff /tmp/inv.spice results/sky130_fd_sc_hd__inv_1.lvs.spice

    diff prints nothing when the files are the same.

Three of those lines are easy to leave out or get wrong:

  • select top cell: drc why reports only on what’s inside the cursor box, and a freshly opened cell has no box worth speaking of. If you skip this line, drc why tells you No errors found. on a cell that has three violations.
  • drc catchup: drc check only queues the re-check, and drc catchup waits for it to finish. If you count before it finishes, the check hasn’t found anything yet, so there’s nothing to count.
  • drc count total, not drc count. Plain drc count prints two lines, Cell sky130_fd_sc_hd__inv_1 has 3 error tiles. and then Total DRC errors found: 0. That second line has the same label as the output of drc count total, but it counts something else: the hierarchical total across subcells, which is always 0 on a flat cell. So if you read the last line of drc count, the number never seems to change.

You should see 0, then 3, then the same three rule names as in Part 4, and no output from diff, which means the flow you typed by hand gave the same netlist as make.

Part 7: Browse the library

How does the library offer more drive strength for the same logic?

  1. Count the cells in the library and list the inverters:

    ls $PDK_ROOT/sky130A/libs.ref/sky130_fd_sc_hd/mag/ | wc -l
    ls $PDK_ROOT/sky130A/libs.ref/sky130_fd_sc_hd/mag/ | grep __inv_

    The first prints 446. The second lists seven inverters: inv_1, inv_2, inv_4, inv_6, inv_8, inv_12 and inv_16. They all have the same logic, and the drive strength goes up from each one to the next.

  2. Open the whole library GDS in KLayout, which is a much better viewer than Magic. make klayout starts it in viewer mode, where it can’t edit anything, so you can browse safely:

    make klayout
  3. Before you look, write down a prediction: is inv_8 taller than inv_1, wider, or both? Then open inv_8 next to inv_1 and see where the factor of eight went.

    It’s only wider. Every cell in the library is 2.72 µm tall, and each step up in drive strength is the same logic with wider transistors (Anatomy of a standard cell), so the extra area has to go sideways.

You should see 446 cells, seven inverters, and an inv_8 that is visibly wider than inv_1.

Part 8: Explore SiliWiz (optional)

This part runs in your browser and needs no setup. It takes about twenty minutes.

  1. Open https://app.siliwiz.com/. SiliWiz describes the interface. Load the inverter from the Preset dropdown above the drawing area to see a finished example.
  2. Open LESSONS (top right) and do the nmos lesson. Drag the cross-section cut line with the blue slider so that it passes through the middle of the poly stripe, and look at the cross-section: substrate, a thin oxide, poly on top, and doped regions on either side.
  3. Now drag the cut so that it misses the poly. The cross-section becomes two doped regions with nothing joining them, because the transistor exists only where the poly crosses.
  4. Move a contact until it hangs off the edge of the diffusion it sits on, and open the CROSS SECTION & DRC tab. SiliWiz’s rules aren’t SKY130’s, but its complaint, “this thing does not surround that thing by enough”, has the same shape as half the errors you’ll fix in Lab 02.
  5. Do the parasitics lesson. Stretch a metal wire longer and watch what happens to the simulated waveform. That’s the argument of Lab 05, made in two minutes.

You should see the transistor appear and disappear in the cross-section as the cut moves on and off the poly, and a DRC complaint as soon as the contact leaves its diffusion.

Finishing

You’re done when make ends in PASS and the W and L you worked out from the .mag file in Part 3 match the W= and L= it prints for both transistors.

Questions:

  • Why does the full rule set find three violations in a cell that has been taped out hundreds of times? Part 4, step 3 answers it.
  • Why does the last line of drc count never seem to change when you switch rule sets? Part 6 answers it.

Submit: nothing for this lab.

Next on the Course path: read Drawing in Magic, Reading DRC errors and The full DRC rule set, then do Lab 02: Break it on purpose.