FOSSlab

Chapter 14 of 14 · about 20 min

For teachers: install and supervise

FOSSLab runs every program inside one container image (IIC-OSIC-TOOLS, with the IHP SG13G2 PDK) through Apptainer, as the user who starts it. Students log into one server; each has a workspace in their home; teachers read every workspace from one window.

Install on the server (Scientific Linux 7, also EL8/9/10)

  1. Copy the fosslab folder to the server and edit install/fosslab.conf: teachers’ user names, groups, image version, tutorial URL.
  2. Run, as root:
    $ sudo ./install/install.sh
    It goes through six steps (each can be run again on its own, e.g. install.sh 40 50):
    StepDoes
    10host packages: points SL7’s and EPEL 7’s repositories to their archives (both are end-of-life), installs Apptainer, ACL tools, X11 forwarding, fonts, and X2Go with XFCE
    20lets normal users run Apptainer (setuid helper, or user namespaces)
    30the image: builds iic-osic-tools_<tag>.sif from Docker Hub (about an hour, ~45 GB free during the build, ~15 GB after), or downloads one built elsewhere with install/build_sif.sh
    40FOSSLab into /opt/fosslab, /etc/fosslab.conf, the PATH, desktop menu entries
    50groups fosslab (students) and fosslab-staff (teachers)
    60the self-test as a teacher: every tool and the full IHP flow
  3. Add the students, from a CSV file (user,Full Name[,password]) or by name:
    $ sudo ./install/add_students.sh students.csv
    New accounts without a password get a random one (in /root/fosslab_passwords.txt), to be changed at the first login.

Supervising

Start fosslab-supervisor (desktop menu: FOSSLab Supervisor). It is the Library Manager plus a Student selector and a Roster:

  • The roster lists every student with their libraries, cells, last change and how many simulations, DRC, LVS and PEX runs pass; private notes per student; export to CSV.
  • Choosing a student shows their workspace, read only: open any view, browse the hierarchy, look at their DRC/LVS reports. Flows you run on their cells write to your own workspace (~/fosslab/.fosslab/supervisor/<student>), never to theirs.
  • Copy to my workspace on any of their cells, to try a fix.

Access works through file ACLs, not passwords: the fosslab-staff group can traverse each student’s home and read ~/fosslab, and new files inherit that. Nothing else in the students’ homes is readable. To let teachers also write (fix a student’s file in place), set FOSSLAB_ACL_MODE=rwX in /etc/fosslab.conf and run install.sh 50.

The reference op-amp library

Everything in chapters 2 to 11 (numbers and screenshots) comes from a reference library that FOSSLab generates, so you can reproduce it, give it to the students, or change it:

$ fosslab example opamp                # library opamp_tut in your workspace
$ fosslab example opamp --solution     # also the exercise's solution (opamp_va2, tb_opamp_va2)

It writes the op-amp (schematic, symbol, layout), the testbenches tb_opamp_ac and tb_opamp_tran with their Simulation Environment setups (outputs and specifications, 15 corners, Monte Carlo), the Verilog-A model opamp_va and its testbench tb_opamp_va. Then compile the model (fosslab va opamp_tut opamp_va) and run what you need.

Sharing it read-only with the students. Generate it without the solution in a workspace of your own (e.g. a course account, or your ~/fosslab) and make that workspace readable by the students’ group. Each student adds one line to their ~/fosslab/fosslab.lib (File > Edit library definitions):

INCLUDE /home/teacher/fosslab/fosslab.lib

The reference library then appears in their Library Manager, read only; Copy cell… gives them an editable copy of any cell.

Changing the example

The generator is /opt/fosslab/app/fosslab/examples/opamp.py: the device sizes (MOS, CC_SIDE), the testbenches’ control blocks (AC_CONTROL, TRAN_CONTROL), the setups with specifications and corners (ac_setup(), tran_setup(), va_setup()) and the model’s parameters. The layout is drawn by examples/opamp_layout.py with IHP’s PCells and reads the sizes from opamp.py; its placement and routing rules are explained at the top of the file. After a change, generate into a new library (--lib opamp_v2) and check it: fosslab drc|lvs opamp_v2 opamp.

All the tutorial’s screenshots and numbers are produced by one script, from scratch, with the real programs: tests/make_tutorial_shots.py (it writes results.json with every number quoted in the chapters). Run it after changing the example to update the pictures.

The exercise: answer key

QuestionExpected answer
1. GBW = 16.17 MHz, FP2 = 30 MHzUGF ≈ 14.1 MHz, PM ≈ 64.4°. The second pole reduces |A| near the UGF, so the gain reaches 1 earlier: with two poles UGF < GBW.
2. CalibrationFP2 = UGF / tan(90° − PM) = 16.17 / tan(29.62°) ≈ 28.4 MHz; GBW = 16.17 · √(1 + (16.17/28.4)²) ≈ 18.6 MHz (GBW_sd scaled to ≈ 0.262 MHz). Model: 16.2 MHz, 59.8° (the small error: the dominant pole’s own phase and the rounding).
3. Monte CarloUGF 16.16 ± 0.18 MHz, PM 59.9 ± 0.29°, offset 1.58 ± 2.04 mV.
4. Spread of PMFrom the GBW spread: UGF moves with GBW while FP2 is fixed, so UGF/FP2 and the PM change. It matches the transistors’ σ (0.30°) here by coincidence of the two mechanisms; a model with correlated GBW and FP2 (both set by the bias current) would be more faithful.
5. Second polep2 ≈ gm6 / (2π (CL + C1)) with the Miller effect, and a right-half-plane zero at gm6/(2π Cc) that also costs phase: the calibrated FP2 lumps both.

The solution model is opamp_va2.va in the solution library (--solution); keep it out of the students’ reach (generate it in a library they cannot include).