Gallery
PDK Devices in 3D
Real layouts of the analog devices in SG13G2 and SKY130, built from each PDK's own device generators and extruded through its layer stack. Drag to rotate, scroll to zoom.
- Recognize the layers a MOSFET, resistor, capacitor, diode, bipolar transistor, inductor and pad are built from.
- Compare the front end (diffusion, poly, contacts) and the back end (metals and vias) of the two PDKs at true scale.
- Explain why the same device looks different in SG13G2 and SKY130 (local interconnect, top metals, MIM position).
- Regenerate every model, or turn your own GDS into a 3D model, with the scripts in
tools/3d/.
Contents
How to read these models
A layout is a stack of two-dimensional drawings, one per mask layer. A chip is three-dimensional: each layer is a slab of material with a bottom and a top. Every model on this page was made in the same three steps.
- Draw the device with the PDK’s own generator. In SG13G2 the IHP KLayout PyCells draw it (the cells you place from KLayout’s library browser) [14, 39]. In SKY130, Magic’s device generators draw it with the
sky130Atechnology from open_pdks [40, 20], and the devices SkyWater ships only as fixed, qualified layouts (bipolar transistors, test inductors) are copied from the PDK’s GDS library [19]. Nothing is drawn by hand. - Extrude every layer through the process stack. Each GDS layer is merged and lifted to its height in the process: SG13G2 from the layer-thickness table of the process specification [15], stacked as in the PDK’s KLayout 2.5D view; SKY130 from SkyWater’s process stack diagram [18], which is also the height table of the Magic technology file.
- Write one 3D file per device in the glTF binary format [50], with one named material per layer, and show it with Google’s
<model-viewer>web component [49].
Everything above the silicon surface is to scale, with no vertical exaggeration. Layer thicknesses, the gaps between layers (the oxide is not drawn, so you see through it) and the via heights are the PDK numbers. Below the surface the drawing is schematic: diffusion is a slab as deep as the trench isolation (400 nm in SG13G2, 120 nm as in Magic for SKY130), and wells, buried layers and the substrate are thin translucent plates so that they do not hide the device. Implant and marker layers that change a region’s doping without adding material (salicide block, thick-oxide marker, the resistor implants) are drawn as thin translucent sheets.
How to use the viewers:
- Rotate: drag with the mouse or one finger. Zoom: click into a viewer first, then use the scroll wheel (so that scrolling down the page never gets stuck in a model), or pinch on a touch screen. Pan: right-drag or drag with two fingers.
- Layers: click a layer in the legend under a viewer to hide or show it; shift-click (or alt-click) shows that layer alone. Hide the top metals to look inside a device.
- Reset view returns to the starting camera. A model loads only when it scrolls into view, so the page stays light.
- The explorer in the next section shows the same models with an explode slider, a vertical scale, top and side views, and both PDKs side by side.
The colours are the same in both PDKs (Metal1 and met1 are both blue, the contacts dark grey, all vias light grey), so that each pair of viewers can be compared at a glance. The left viewer is always SG13G2 and the right one SKY130; on a phone they stack.
Table 1. The two layer stacks as drawn: z is the bottom of each layer in µm, for SG13G2 above the bottom of the 400 nm trench-isolation level (as in the PDK's KLayout 2.5D view), for SKY130 as in SkyWater's process stack diagram. Sources: SG13G2 process specification, section 2.16; SkyWater SKY130 documentation.
| Level | SG13G2 layer | z (µm) | Thickness | SKY130 layer | z (µm) | Thickness |
|---|---|---|---|---|---|---|
| Silicon surface | top of Activ | 0.40 | (STI 0.40) | top of diff | 0.326 | (0.12) |
| Gate | GatPoly | 0.40 | 0.16 | poly | 0.326 | 0.18 |
| Local interconnect | – | – | – | li1 | 0.936 | 0.10 |
| Metal 1 | Metal1 | 1.04 | 0.42 | met1 | 1.376 | 0.36 |
| Metal 2 | Metal2 | 2.00 | 0.49 | met2 | 2.006 | 0.36 |
| Metal 3 | Metal3 | 3.03 | 0.49 | met3 | 2.786 | 0.845 |
| Metal 4 | Metal4 | 4.06 | 0.49 | met4 | 4.021 | 0.845 |
| Metal 5 | Metal5 | 5.09 | 0.49 | met5 | 5.371 | 1.26 |
| MIM top plate | MIM (above Metal5) | 5.62 | 0.15 | capm (above met3) | 3.651 | 0.10 |
| Thick metal 1 | TopMetal1 | 6.43 | 2.0 | – | – | – |
| Thick metal 2 | TopMetal2 | 11.23 | 3.0 | – | – | – |
| Passivation | oxide + nitride | 14.23 | 1.5 + 0.4 | oxide + nitride | 6.631 | 0.09 + 0.54 |
Explore
The explorer takes the devices of this page apart. Show one PDK, or both side by side at the same scale; pull the layers apart with Explode; stretch the heights with Vertical scale; look straight down with Top (the layout as you draw it) or from the front with Side (the stack in cross-section); and switch layers on and off in the panel, which also lists the height of each one. It loads the same models as the viewers further down, and every pair of viewers has a button that opens its device here.
The 3D view loads when you scroll to it (it needs JavaScript and WebGL).
Drag to rotate; right-drag or two fingers to pan; click into the view, then scroll or pinch to zoom. In the layer panel, shift-click (or alt-click) a layer to show it alone.
Try it · Five things to look for
- The full metal stack, both PDKs, Side. The SG13G2 stack is more than twice as tall as SKY130’s, and most of the difference is its two thick top metals. Switch the explode slider up and count the levels.
- Core NMOS, Top, then hide the metals. What is left is the transistor itself: the gate where poly crosses active area. Compare the gate lengths, 0.13 and 0.15 µm, with the contacts on either side.
- Multi-finger NMOS, Explode. Each level separates: diffusion, poly fingers, contacts, (li1 in SKY130) and Metal1. Count the source and drain columns.
- MIM capacitor, Side at 4× vertical scale. Find the bottom plate, the thin top plate just above it and the via that contacts it from the next metal; note at which metal level each PDK puts its capacitor.
- Inductor, both PDKs. At the same scale the SKY130 test coil is the larger device, but it lies in the thin middle metals; the SG13G2 spiral is drawn in the 3 µm TopMetal2, high above the substrate.
The full metal stack
Start with the whole back end of line: a contact on diffusion and a tower of vias up to the top metal. The SG13G2 tower comes from the PyCell via_stack (from Activ to TopMetal2, with 2 × 2 via arrays up to Metal5); the SKY130 tower was painted in Magic as n-diffusion, then each contact type from ndiffc up to via4, and Magic generated the via arrays from them.
Footprint 2 × 2 µm, total height 14.2 µm. TopMetal1 (2 µm) and TopMetal2 (3 µm) make up more than a third of the height.
Footprint 2.4 × 2.4 µm, height 6.6 µm to the top of met5. Note li1, the thin slate plate just above the contacts.
The two back ends are built for different jobs. SG13G2 has five thin aluminium metals of about 0.5 µm and two thick top metals of 2 and 3 µm, made for low-loss RF wiring and inductors: TopMetal2 has a sheet resistance of 11 mΩ/□ against 88 mΩ/□ for Metal2 to Metal5 [15]. SKY130 has a local-interconnect layer, li1, and five metals that get thicker towards the top (0.36, 0.36, 0.845, 0.845 and 1.26 µm). li1 is titanium nitride, only 0.1 µm thick, with a sheet resistance of 12.8 Ω/□ (a hundred times that of met1) [18, 20], so it is used for short connections inside a cell, never for long wires. In SKY130 every contact lands on li1 first, and a second via (mcon) reaches met1. That extra level is why the SKY130 devices below all carry a slate-grey layer between the contacts and the blue met1.
Hide the metals one by one from the top and look at the vias: in SG13G2 each via level up to Metal5 is a 2 × 2 array of small vias, then single, much larger TopVia1 and TopVia2 cuts follow. A via is a resistor too, from about 9 Ω for Via1 to 1.1 Ω for TopVia2 [15]; Chapter 15 shows why wide wires need arrays of them.
The core NMOS transistor
The most important device of the book, drawn at minimum length: a single gate finger W = 1 µm across a rectangle of active area. The gate poly (red) crosses the active area (green, n+); the two halves of the active area on either side are the source and the drain, each with a column of contacts up to Metal1.
W = 1 µm, L = 0.13 µm, 1 finger, as the PyCell draws it (no guard ring).
W = 1 µm, L = 0.15 µm, 1 finger, with the substrate-tap guard ring Magic adds by default.
The gate is only 160 nm (SG13G2) or 180 nm (SKY130) of polysilicon on a gate oxide a few nanometres thick: 2.65 nm under an SG13G2 NMOS [15]. The oxide is far too thin to see at this scale, which is the point: the whole transistor action happens in a sheet of silicon under the red bar. Look for three differences between the two PDKs.
- The guard ring. Magic surrounds the device with a p+ substrate-tap ring by default; the IHP PyCell draws the bare transistor unless you ask for a ring (parameter
guardRingType). Real analog layouts almost always add the ring (Section 15.5). - The local interconnect. In SKY130 the contacts land on li1, and only then does
mconreach met1. - The gate contact. The IHP PyCell leaves the gate uncontacted: the poly simply extends 0.18 µm past the active area and you connect it in your own routing. Magic adds a poly head with a contact above and below the device (parameters
topcandbotc).
Chapter 4 explains what happens in the silicon under the gate.
Multi-finger NMOS
Wide transistors are folded into several parallel fingers that share their source and drain diffusions. This keeps the device compact, reduces the drain junction capacitance (each inner drain serves two fingers) and lowers the gate resistance, because each poly finger is short.
W = 8 µm total (4 fingers of 2 µm), L = 0.5 µm.
W = 8 µm total (4 fingers of 2 µm), L = 0.5 µm.
The source and drain columns alternate (S D S D S), so the two outer columns are both sources and the device is symmetric. This is the first step towards the interdigitated and common-centroid layouts of Section 15.4. In SKY130, recall from Chapter 6 that the models are binned by finger width, so the finger width you choose here also chooses the model bin.
PMOS in its n-well
A PMOS transistor needs an n-type body, so it sits in an n-well (translucent violet), and the well must be tied to the supply through an n+ contact. Its source and drain are p+ diffusion (ochre).
W = 2 µm (2 fingers of 1 µm), L = 0.13 µm, with the n-well guard ring (guardRingType = nwell).
W = 2 µm (2 fingers of 1 µm), L = 0.15 µm, with Magic's n-well tap ring.
Here the guard ring is not optional: the green n+ ring is the body contact. Hide NWell / nwell to see that the well is simply a larger rectangle under the whole device. In SG13G2 the PyCell also draws nBuLay, an n+ buried layer, under the whole well. A PMOS in its own well is the reason a CMOS layout has to group PMOS devices together and keep a spacing to NMOS devices (Section 3.6).
Thick-oxide (high-voltage) MOSFET
Both PDKs offer a second gate oxide for higher voltages: SG13G2’s 3.3 V devices and SKY130’s 5 V devices. Geometrically they are ordinary MOSFETs with a longer minimum length, drawn inside a marker layer that tells the fab to grow the thicker oxide (translucent orange sheet).
W = 2 µm, L = 0.45 µm (the minimum for this device), ThickGateOx marker.
W = 2 µm, L = 0.5 µm (the minimum), inside the hvi thick-oxide marker; 5 V gate, 10.5 V drain.
The SG13G2 thick oxide is 7.5 nm under an NMOS, about three times the 2.65 nm of the core device [15]. A thicker oxide tolerates a larger gate voltage but gives less gate capacitance per area, so these devices are slower and have a higher threshold. The circuits in this book use the core devices; thick-oxide devices are for I/O circuits and blocks that must handle a higher supply.
Polysilicon resistors
A poly resistor is a strip of gate polysilicon with a contact head at each end. What makes the difference between the resistor types is the doping of the strip and whether its surface is salicided (covered with a low-resistance metal silicide). The salicide is blocked over the resistor body by a marker layer (the light-blue translucent sheet).
From left: rsil, rppd and rhigh, each W = 0.5 µm, L = 3 µm (PyCells with Calculate = R).
From left: res_high_po_0p35 and res_xhigh_po_0p35, each W = 0.35 µm, L = 3 µm.
Table 2. Sheet resistance of the poly resistors shown above.
| Resistor | Material | Sheet resistance | Source |
|---|---|---|---|
SG13G2 rsil |
salicided n+ poly | 7.0 Ω/□ | [15] |
SG13G2 rppd |
unsalicided p+ poly | 260 Ω/□ | [15] |
SG13G2 rhigh |
unsalicided, partly compensated poly | 1360 Ω/□ | [15] |
SKY130 res_high_po |
p+ poly with resistor implant | 319.8 Ω/□ | [20] |
SKY130 res_xhigh_po |
poly with the ultra-high resistor implant | 2000 Ω/□ | [20] |
A 3 µm long, 0.5 µm wide strip is 6 squares, so the three SG13G2 resistors above are about 42 Ω, 1.6 kΩ and 8.2 kΩ plus their contact resistance. The SKY130 precision resistors only come in fixed widths (0.35, 0.69, 1.41, 2.85 and 5.73 µm), and their heads use long slot contacts, which you can see in the model. Which type to use for a bias resistor, and how to make two resistors match, is the subject of Section 15.6.
MIM capacitor
A metal-insulator-metal (MIM) capacitor is a thin top plate placed over a metal layer with a very thin dielectric in between. The bottom plate is the metal itself; the top plate is contacted from the metal above.
5 × 5 µm, between Metal5 (bottom plate) and TopMetal1 (top contact). Hide TopMetal1 to see the plate.
5 × 5 µm, between met3 (bottom plate) and capm, contacted from met4.
The SG13G2 MIM dielectric is 40 nm thick and gives 1.5 fF/µm² [15]; the SKY130 cap_mim_m3_1 gives about 2 fF/µm² [20]. The 5 × 5 µm capacitors above are therefore roughly 38 fF and 50 fF. Notice where each one sits: high in the stack, far from the substrate, which keeps the parasitic capacitance of the bottom plate small. A 2 pF capacitor, the load of the course OTA (Chapter 12), would need about 1330 µm² of SG13G2 MIM, a square 36 µm on a side.
MOS varactor
A varactor is a voltage-controlled capacitor. The MOS varactor is a MOSFET-like structure in an n-well whose gate capacitance changes with the gate-to-well voltage; it tunes voltage-controlled oscillators.
One 3.74 × 0.3 µm cell, thick gate oxide.
W = 2 µm, L = 0.5 µm, 1 finger.
Unlike a MOSFET, both sides of the gate are n+ in an n-well, so there is no channel of the opposite type: the device moves between accumulation and depletion. In the PDK model the SG13G2 S-Varicap moves from about 2.5 to 5.8 fF per µm² of gate between −3 V and +3 V, a tuning range of 2.3 : 1 (the specification’s 23 to 39.5 fF/µm² is normalized to its test structure, not to the gate area [15]; Chapter 26 compares the varactors of both PDKs). This book’s circuits do not use varactors; they are here because they complete the PDKs’ device list.
Diode
The simplest junction device: an n+ diffusion in the p-type substrate (or p-well), with its anode contacted through a p+ tap.
1.5 × 1.5 µm n+/p-substrate diode (the PDK's antenna diode cell).
1.5 × 1.5 µm n+/p-well diode with its p+ anode ring.
The same structure is hidden in every NMOS: the source and drain are n+/p diodes to the substrate, and their capacitance is part of every node capacitance in Chapter 9. As a separate cell the diode protects gates from charge collected during fabrication (the antenna effect, Section 16.6).
Bipolar transistors
Both PDKs contain bipolar transistors, but of very different kinds. SG13G2 is a BiCMOS process: its npn13G2 is a silicon-germanium heterojunction bipolar transistor (HBT) with a transit frequency of 300 GHz and a maximum oscillation frequency of 500 GHz [15]. SKY130 offers bipolars that are built from the CMOS layers (wells and diffusions), with fixed, characterized layouts. The vertical PNP of both PDKs is the workhorse of bandgap references (Section 14.3).
NPN transistors
One emitter, 0.07 × 0.9 µm (emitter window EmWind, pink). Hide Metal2, Via1 and Metal1 to see it.
Fixed layout, 1 × 1 µm emitter, copied from sky130_fd_pr.
The HBT model shows only the drawn layers. The parts that make it fast, the epitaxial SiGe base and the polysilicon emitter that fills the 70 nm wide emitter window, are formed by the process from these drawings, so the pink stripe stands in for the emitter. Its narrowness is the point: a short distance for the carriers, hence the high transit frequency. The SKY130 NPN is built from wells and diffusions only; it is much larger, and its concentric rings (emitter, base, collector, substrate) show the classic vertical bipolar structure.
PNP transistors
W = L = 2 µm emitter: p+ emitter, n-well base, p-substrate collector.
Fixed layout, 3.4 × 3.4 µm emitter, copied from sky130_fd_pr.
These are vertical PNPs: the emitter is p+ diffusion, the base is the n-well and the collector is the p-type substrate itself, so the collector is always at ground. That restriction is exactly what a bandgap reference needs, which is why both PDKs characterize this “parasitic” device carefully.
Inductor
An on-chip spiral inductor is a wide metal track wound into a few turns, with an underpass on a lower metal to bring the inner end out.
Two turns, track 4 µm wide, spacing 2.1 µm, in TopMetal2 (3 µm thick), underpass in TopMetal1.
SkyWater's fixed test coil, about 150 µm across, in met3 with a met2 underpass.
Here the difference between the back ends is decisive. SG13G2 has a parametric inductor PyCell on its two thick top metals, 2 and 3 µm of aluminium far above the lossy substrate. SKY130 provides only fixed test coils in its thinner middle metals. Both models are at true scale, so the 3 µm TopMetal2 track looks thin only because the inductor is so wide. This book’s circuits do not use inductors.
Bond pad
A pad is where a bond wire or a probe needle touches the chip: a large square of top metal under an opening in the passivation, the protective glass layer over the whole chip (here a translucent sheet with the window cut out).
60 × 60 µm square pad, TopMetal2 down to Metal3 with full via arrays, passivation opening on top.
Bare pad from the Efabless I/O library, about 60 × 70 µm, met5 and met4.
Only the inductor is larger: one pad covers more area than all the transistors, resistors and capacitors on this page together. The stack of metals and vias under it spreads the force of wire bonding. The pad frame, the electrostatic-discharge protection that must sit next to every pad, and the bonding itself are covered in Chapter 17.
Make your own 3D models
Everything on this page is regenerated by one script, which runs in IIC-OSIC-TOOLS [35] (or starts the container with Docker if you call it from outside):
tools/3d/make_devices.sh # GDS -> tools/3d/gds/, 3D models -> assets/3d/ (about 1 minute)
It calls the SG13G2 PyCells (tools/3d/gen_sg13g2.py), Magic’s SKY130 generators (tools/3d/gen_sky130.tcl), copies the fixed SKY130 layouts (tools/3d/fixed_sky130.py) and converts each GDS with tools/3d/gds2glb.py. The converter works on any layout, for example the OTA you lay out in Chapter 15:
python3 tools/3d/gds2glb.py my_ota.gds my_ota.glb --stack sg13g2 # whole top cell
python3 tools/3d/gds2glb.py my_ota.gds core.glb --stack sky130 --clip 0 0 40 30 --skip Substrate
python3 tools/3d/preview.py my_ota.glb my_ota.png --orbit 35 60 # PNG check, no browser
Options: --cell (which cell), --zscale (vertical exaggeration, default 1), --clip X0 Y0 X1 Y1 (a window in µm), --only and --skip (layers by legend name), --margin (substrate margin). The layer stacks live in tools/3d/stacks.py, one table per PDK with each layer’s GDS number, a Boolean expression (e.g. Cont or ActivAll & pSD), its bottom and top height, colour and opacity; add your own as a Python file. Open the .glb in any glTF viewer, in Blender, or on a page of your own with <model-viewer>.
Try it · Lab 8
In Lab 8 you lay out and verify a small circuit. Convert your final GDS with
gds2glb.pyand look at it from below the metals: it is the quickest way to spot a missing via or a wire that ended on the wrong layer.
- Every device is a few patterned layers: diffusion and poly in the silicon, contacts, then a stack of metals and vias. The generators of both PDKs draw them for you.
- SKY130 routes through a thin, resistive local interconnect (li1) before met1; SG13G2 contacts go straight to Metal1.
- SG13G2’s back end ends in two thick top metals (2 and 3 µm) for inductors and low-loss wiring and holds its MIM capacitor above Metal5; SKY130’s MIM sits above met3.
- Resistor types differ by doping and salicide, not by shape: 7 Ω/□ to 2000 Ω/□ from the same poly layer.
- The bipolar devices differ most: a 300 GHz SiGe HBT in SG13G2, well-and-diffusion bipolars in SKY130, and a vertical substrate PNP in both for references.
Further reading
- The SG13G2 process specification [15] gives every layer thickness and device parameter used here; the SkyWater documentation [18] has the SKY130 process stack diagram and a page per device.
- The KLayout documentation [39] describes its 2.5D view, which shows a layout in 3D inside the layout editor, and both PDKs ship a 2.5D script for it.
- Magic [40] can also show the layout it is editing in 3D, from the same height table that was used here for SKY130.