From schematic to cross-section
So far in AD102 you have drawn rectangles and called them components. This page shows what those rectangles look like when you slice through the chip.
Everything up to now has been in plan view: you drew a shape on a mask layer and read a number off a simulator. That’s how layout is done, and it’s enough to design with. A chip, though, is a stack of films a few micrometres tall, and every surprise in this course came from that third dimension:
- The resistor had end resistance (378.2448 Ω of it) because current has to climb down from metal into poly and back up again.
- The capacitor had fringe capacitance because field lines leave the edge of a plate.
- The wire had a parasitic to the substrate because it is a bar suspended over a plane.
- The inductor was impractical because its magnetic field goes into the silicon underneath it.
Each of those is a fact about the side view, which is what this page is about.
The layer stack
Ohms per square started from
\[R = \rho\frac{L}{Wt} = \left(\frac{\rho}{t}\right)\frac{L}{W} = R_\square\frac{L}{W}\]
and treated the film thickness \(t\) as a number that the foundry fixes and that you never need to know. The foundry does fix it, but it also writes it down, in the technology file Magic uses to draw cross-sections, and that file has been on your disk since IC101. Its height lines give, for each layer, how far the bottom of the layer sits above the wafer surface and how thick the layer is, both in µm. Here are the lines for the main layers (the file also lists fill, contact, via and MIM layers):
height dnwell -0.1 0.1
height nwell,pwell 0.0 0.2062
height alldiff 0.2062 0.12
height allpoly 0.3262 0.18
height allli 0.9361 0.10
height allm1 1.3761 0.36
height allm2 2.0061 0.36
height allm3 2.7861 0.845
height allm4 4.0211 0.845
height allm5 5.3711 1.26
height mrdl 7.2611 3.0
Those eleven lines describe the main stack of the chip. It helps to read them like the floors of a building, each with the height it starts at and how tall it is:
| Layer | Bottom | Thickness |
|---|---|---|
| deep n-well | −0.1 µm | 0.1 µm |
| n-well / p-well | 0.0 µm | 0.2062 µm |
| diffusion | 0.2062 µm | 0.12 µm |
| poly | 0.3262 µm | 0.18 µm |
| local interconnect (li) | 0.9361 µm | 0.10 µm |
| metal1 | 1.3761 µm | 0.36 µm |
| metal2 | 2.0061 µm | 0.36 µm |
| metal3 | 2.7861 µm | 0.845 µm |
| metal4 | 4.0211 µm | 0.845 µm |
| metal5 | 5.3711 µm | 1.26 µm |
| redistribution metal | 7.2611 µm | 3.0 µm |
The top of metal5 is 6.6311 µm above the silicon, so the wiring stack up to metal5 is about as thick as a red blood cell is wide. (The redistribution metal above it reaches 10.2611 µm, and the substrate underneath isn’t counted at all.) Your 100 kΩ resistor was 314.046460 µm long, which is forty-seven times the height of that stack. You’ll print this stack yourself in the capstone, Part 1.
From sheet resistance to resistivity
In Lab 01 you measured the sheet resistance of high-sheet poly yourself, 317.2198 Ω/□, and the tech file says that poly is 0.18 µm thick, so its resistivity is
\[\rho = R_\square \, t = 317.2198 \times 0.18\times10^{-6} = 5.7100\times10^{-5}\ \Omega\,\text{m} = 5.7100\ \text{m}\Omega\,\text{cm}\]
A sheet resistance depends on how thick the film is, but this number doesn’t: it’s a material property, the resistivity of the p+ doped polysilicon that SkyWater deposits, and you got it from a resistance you measured and a number you read out of a file.
Do the same for res_xhigh_po, which measured 2118.7619 Ω/□ on the same 0.18 µm film:
\[\rho = 3.8138\times10^{-4}\ \Omega\,\text{m} = 38.1377\ \text{m}\Omega\,\text{cm}\]
That’s the same layer at the same thickness, with 6.7 times the resistivity. Resistors on a chip said that res_high_po and res_xhigh_po differ only in implant dose, and the factor of 6.7 is what that difference in dose does to the material: less dopant means fewer carriers, and fewer carriers mean a higher resistivity.
To give those numbers a scale, here they are next to two metals:
| Material | \(\rho\) | vs copper |
|---|---|---|
| copper | 1.68×10⁻⁸ Ω m | 1× |
| SKY130 metal1 (0.125 Ω/□ × 0.36 µm) | 4.50×10⁻⁸ Ω m | 2.7× |
res_high_po |
5.71×10⁻⁵ Ω m | 3 399× |
res_xhigh_po |
3.81×10⁻⁴ Ω m | 22 701× |
This is why on-chip resistors are made of poly rather than metal. A resistor’s job is to resist, and the wiring metal conducts more than a thousand times too well for that. To get 10 kΩ from metal1 you would need \(10000/0.125 = 80\,000\) squares, which at the 1 µm width of the course’s poly resistors is 80 millimetres of wire on a die only three millimetres across.
It’s worth making a habit of this conversion. Whenever a table gives you Ω/□, you’re one multiplication away from a material property, and material properties tell you whether a number is plausible. A resistivity of 5.7 mΩ·cm is normal for a doped semiconductor, so if your arithmetic had produced 5.7 Ω·cm or 5.7 µΩ·cm, you would know something had gone wrong.
In the capstone, Part 2 you draw a resistor, a capacitor and a wire in SiliWiz and look at the cross-section of each one, and Part 3 then has you size a filter and defend its area.
The three questions AD102 answers
AD102 set out to answer three questions:
- How do you fabricate a resistor at the micro scale? You make it from a strip of doped poly, 0.18 µm thick, with \(\rho = 5.71\) mΩ·cm. Its value is \(R_\square L/W\) plus a fixed end resistance that you can’t avoid, so you buy ohms with length.
- A capacitor? You make it from two plates and a gap: metal3 with a thin dielectric above it for a MIM, a gate oxide for a MOS capacitor, or interleaved wire fingers for a VPP. A linear capacitor gives you at best about 2 fF per µm², so a picofarad needs a plate 22 µm on a side.
- An inductor? Mostly, you don’t make one. A metal spiral gets you single-digit nanohenries at a \(Q\) in the teens and costs tens of thousands of square micrometres, and below about 1 GHz the alternatives do the job better.
All three answers rest on one idea: a component value is a geometry, and geometry costs area. The habit that follows from it is to ask, of any analog schematic anyone shows you, what shape each part is and what that shape costs.
Where you go next
AD103 — Nonlinear Circuits is the next course. It asks the same question (what is this component, physically?) about devices whose behaviour is no longer linear. On Capacitors on a chip you saw the value of the MOS capacitor move by 3.57× with voltage. That was a transistor turning on, and AD103 opens by explaining what you were looking at. AD103 is also where XSchem becomes your main tool, and Lab 03, Part 8 was your introduction to it.
AD104 — Layout is where the cross-section turns into geometry you draw yourself. You draw these devices in Magic, on the real stack in the table above, and then prove two things about them: with DRC, that a fab could make them, and with LVS, that what you drew is the circuit you meant. Two things from this course come back there. The unit-device and common-centroid rules from Matching and Pelgrom’s law become rectangles that you draw, and parasitic extraction reads your finished layout and hands back a netlist that includes every capacitor described on Parasitic capacitance.
Next on the Course path: Capstone: Defend a floor plan.