Appendix C
SG13G2 RF Device Reference
The SiGe HBTs, RF MOSFETs, capacitors, inductors, varactors, lines and metal stack of IHP's SG13G2 process, with the numbers the course measured on them. Process-specification values come from IHP's documentation [200, 191]. Every other number was simulated in IIC-OSIC-TOOLS 2026.10 (typical corner, 27 °C) or electromagnetically simulated with Palace, in the chapter linked beside it.
Contents
C.1 Back end and substrate
| Layer | Thickness | Sheet resistance | Remark |
|---|---|---|---|
| Metal1 | 0.42 µm | 110 mΩ/□ | |
| Metal2–Metal5 | 0.49 µm | 88 mΩ/□ | Metal5 is the MIM capacitor’s bottom plate |
| TopMetal1 | 2.0 µm | 18 mΩ/□ | inductor underpasses, MIM top-plate connection |
| TopMetal2 | 3.0 µm | 11 mΩ/□ | 11.23 µm above the silicon; spirals and transmission lines |
| Substrate | about 50 Ω·cm (2 S/m) | lossy; sets the inductors’ substrate loss |
The stack is drawn to scale beside SKY130’s in Figure 1.5. A 10 µm wide, 1 mm long TopMetal2 line has 1.1 Ω of resistance at DC.
C.2 SiGe heterojunction bipolar transistors
SG13G2 has three npn HBTs. npn13G2 is the fast one, with emitter fingers of 0.07 µm × 0.9 µm (Nx fingers per device). npn13G2l takes a longer emitter, set by its length le. npn13G2v trades speed for breakdown voltage with a 0.12 µm wide emitter. Chapter 6 drew the last two with le = 1 µm; current densities below are per emitter area.
| npn13G2 | npn13G2l | npn13G2v | |
|---|---|---|---|
| fT / fmax, process spec | 300–350 / 400–450 GHz | 280–330 / 380–430 GHz | 90–120 / 280–330 GHz |
| BVCEO / BVCBO, spec | 1.6 / 4.8 V (BVCEO 1.4–1.6 V) | 1.6 / 4.8 V | 2.5 / 8.5 V (BVCEO 2.2–2.5 V) |
| Peak fT, simulated (JC) | 353 GHz (42.8 mA/µm²) | 352 GHz (43.6 mA/µm²) | 104 GHz (3.51 mA/µm²) |
| Peak fmax, simulated (JC) | 456 GHz (38.1 mA/µm²) | 457 GHz (39.3 mA/µm²) | 265 GHz (3.51 mA/µm²) |
| Min. NFmin at 2.4 GHz (JC) | 0.224 dB (0.552 mA/µm²) | 0.211 dB (0.531 mA/µm²) | 0.22 dB (0.369 mA/µm²) |
| Min. NFmin at 50 GHz (JC) | 1.33 dB (4.28 mA/µm²) | 1.17 dB (5.56 mA/µm²) | 1.63 dB (1.46 mA/µm²) |
| Current gain β at peak fT | 363 | 282 | 697 |
Simulated rows: Nx = 4 at VCE = 1.2 V, fT and fmax extrapolated from 40 GHz (Chapter 6). The npn13G2’s peak fT moves little with VCE (338, 353 and 357 GHz at 0.8, 1.2 and 1.5 V), and its model card gives a base resistance rb = rbx + rbi = 28.9 Ω for Nx = 4. This value is pessimistic, because rbi falls at high current: Eq. (6.9) with it predicts an fmax of 257 GHz, against 456 GHz simulated. At 50 GHz an npn13G2 at its fT peak (Nx = 4, 10.8 mA) still offers Gmax = 15.8 dB (MSG); at 2.44 GHz it offers 29.0 dB (Table 1.3).
Breakdown sets the supply of a power stage. For +10 dBm from a class-A or class-B stage that may swing to BVCEO, an npn13G2 needs VCC = 0.8 V and RL = 32 Ω, and an npn13G2v 1.25 V and 78.1 Ω. With a low-impedance base the limit rises towards BVCBO: 2.4 V and 288 Ω for the npn13G2 (Table 6.1). Chapters 16 and 17 design with these limits.
Simulating HBTs. The models are VBIC (cornerHBT.lib, section hbt_typ) and need no OSDI plugin in ngspice. A single instance with a large Nx can stop a transient from converging; several Nx = 4 instances in parallel work. Xyce’s harmonic balance stops converging at high drive, so Chapters 16 and 17 use ngspice transients with an FFT for large-signal results (Appendix B).
C.3 MOSFETs at radio frequencies
| sg13_lv_nmos (1.2 V) | sg13_hv_nmos (3.3 V) | |
|---|---|---|
| Gate oxide | 2.45–2.65 nm | 7.3–7.5 nm |
| Peak fT at VDS = VDD/2 (JD) | 75.7 GHz (0.336 mA/µm) | 16.0 GHz (0.241 mA/µm, L = 0.45 µm) |
| Peak fT at VDS = VDD (JD) | 89.5 GHz (0.424 mA/µm) | |
| Peak fmax (JD) | 89.7 GHz (0.187 mA/µm) | 71.6 GHz (0.0742 mA/µm) |
| Gate resistance, 20 µm in 2 µm fingers, 2 cells | Rg = 25.5 Ω | |
| Min. NFmin at 2.4 GHz (JD) | 0.200 dB (0.153 mA/µm) | |
| Min. NFmin at 50 GHz (JD) | 3.55 dB (0.123 mA/µm) | |
| BVDSS, typical (minimum) | 2.7 V (2.0 V) | 6.1 V (5.3 V) |
Devices 20 µm wide in 2 µm fingers, L = 0.13 µm unless stated (Table 6.2). The pMOS devices sg13_lv_pmos and sg13_hv_pmos share the oxides.
Turn on the RF model and build wide devices from several cells. With rfmode=1 the PSP 103 models add gate resistance and non-quasi-static effects. Without it, SG13G2’s fmax comes out about 2.4 times too high (216 GHz instead of about 89 GHz). The gate resistance has a fixed part, RGO = 40 Ω per instance, divided by the multiplier m. A 200 µm LNA device drawn as one instance with ng=100 therefore carries the full 40 Ω, comparable to the 50 Ω source it should match. Split it into cells with m > 1, as Chapter 9 does (Section 6.2). Xyce reproduces the same gate resistance with rfmode=1 (46.9 Ω against ngspice’s 48.9 Ω for the test device), so do not add Rg by hand.
At 2.4 GHz, Γopt of the 20 µm device is 0.981 ∠ 2.07°, close to an open circuit. At 50 GHz it is 0.670 ∠ 38.7° (Zopt = 68.3 + j104 Ω), where the HBT is the better low-noise device. Table 6.3 gives the course’s choice of device for each block at both frequencies.
C.4 Capacitors
The MIM capacitor (cap_cmim, and cap_rfcmim with its RF parasitics) has 1.5 fF/µm² between a Metal5 bottom plate and a thin top plate tied to TopMetal1. Its Q falls quickly with size and frequency, and a wide feed helps (Chapter 7):
| Capacitor (2 µm feed unless stated) | C | Q at 2.44 GHz | Q at 4.88 GHz | SRF |
|---|---|---|---|---|
| cap_rfcmim 10 µm × 10 µm | 0.152 pF | 760 | 373 | above 100 GHz |
| cap_rfcmim 20 µm × 20 µm | 0.603 pF | 180 | 86.9 | 48.2 GHz |
| cap_rfcmim 30 µm × 30 µm | 1.35 pF | 75.2 | 35.6 | 28.5 GHz |
| cap_rfcmim 30 µm × 30 µm, 10 µm feed | 1.35 pF | 191 | 87.3 | 33.3 GHz |
| cap_rfcmim 40 µm × 40 µm | 2.41 pF | 39.8 | 18.2 | 19.2 GHz |
| cap_rfcmim 75 µm × 75 µm | 8.45 pF | 8.83 | 2.88 | 7.94 GHz |
| cap_cmomi 30 µm × 30 µm (MOM) | 1.18 pF | 134 | 66.1 | 44.2 GHz |
Above a picofarad or so, several small capacitors in parallel, each with its own wide feed, beat one large plate.
C.5 Spiral inductors
Chapter 7’s inductor family is drawn on TopMetal2 with a TopMetal1 underpass and characterized with Palace through IHP’s gds2palace flow [197, 198] (Table 7.2). Single-ended values (differential in brackets):
| L1 | L2 | L3 | L4 | |
|---|---|---|---|---|
| Turns, dout, width | 1, 200 µm, 10 µm | 2, 150 µm, 8 µm | 3, 170 µm, 8 µm | 4, 210 µm, 6 µm |
| L at 1 GHz (EM) | 0.439 nH | 0.9 nH | 1.92 nH | 4.73 nH |
| RDC (EM) | 0.724 Ω | 1.14 Ω | 1.74 Ω | 3.41 Ω |
| Q at 2.44 GHz | 7.91 | 8.91 | 10.6 | 11.3 |
| Q at 4.88 GHz | 13.7 (14) | 15.1 (15.6) | 16 (17.7) | 10.8 (15.6) |
| Peak Q, at | 22.3, 17 GHz | 21, 11.8 GHz | 17.1, 6.75 GHz | 12.4, 3.5 GHz |
| SRF | above 50 GHz | 33 (39.6) GHz | 16.9 (18.7) GHz | 8.12 (8.65) GHz |
A π model with a skin-effect branch fits each spiral from DC to about 0.8 × SRF (Table 7.3). tools/figs/_ch07_inductor.py interpolates it for any inductance from 0.424 to 4.94 nH (inductor_lines(name, n1, n2, L_nH)), and every later chapter uses it.
The flow was checked against an independent HFSS characterization of the 4 nH, three-turn octagonal tank inductor of a taped-out SG13G2 2.4 GHz PLL [106, 207]. At 2.45 GHz Palace gives Ldiff = 3.69 nH (−8 %) and Qdiff = 16.3 (−3 %); the differential SRF is 10.8 GHz (+8 %). Pass model="V4" to use that spiral.
C.6 Varactors and switched capacitors
| Element | Range | Q | Remark |
|---|---|---|---|
| sg13_hv_svaricap, process spec | about 23–40 fF/µm² | 43–62 at 15.8 GHz | accumulation-mode MOS varactor in an n-well |
| svaricap, L = 0.8 µm, 10 fingers (W = 9.74 µm) | 89.6–209 fF over ±1.5 V | 37.9 minimum at 4.88 GHz | Cmax/Cmin = 2.34; differential, between the two gates |
| Switched MIM, switch W = 80 µm | 301 fF on | 12.9 at 4.88 GHz | Con/Coff = 11.2 |
The varactor is a mosvar device that Xyce’s single-plugin build cannot load: simulate tuning curves in ngspice (Chapter 7, Chapter 11).
C.7 Transformers, transmission lines and pads
| Element | Key numbers | Source |
|---|---|---|
| Stacked transformer T1 (1 turn, TopMetal2 over TopMetal1) | 0.437 nH, k = 0.72; minimum insertion loss 1.78 dB at 2.44 GHz, 1.01 dB at 4.88 GHz, 0.372 dB near 38 GHz; SRF 42.2 GHz | Chapter 7 |
| TopMetal2 microstrip, 16 µm over Metal1 | Z0 = 50.7 Ω at 4.88 GHz, 47.8 Ω at 50 GHz; εeff = 3.73 at 50 GHz; loss 0.283, 0.325 and 0.583 dB/mm at 2.44, 4.88 and 50 GHz | Chapter 7 |
| GSG signal pad, TopMetal2 over a Metal1 shield | 80 µm square (150 µm pitch): 37.2 fF at 1–2.44 GHz, 37.7 fF at 50 GHz, 56 % above its parallel-plate value; 60 µm square: 23.8 fF. The PDK’s stacked bondpad is about 123 fF (hand estimate) |
Chapter 18 |
| TopMetal2 microstrip, 15 µm, 880 µm long, over Metal1 | openEMS and Palace agree within 0.2 dB (Palace with first-order elements) | Chapter 2 |
The wavelength on a TopMetal2 line is 5.61 cm at 2.4 GHz and 3.1 mm at 50 GHz. A wire of a few hundred micrometres is lumped at the course radio’s frequency, but at millimetre waves it must be designed as a transmission line.
C.8 Resistors
| Resistor | Sheet resistance | Typical use |
|---|---|---|
rsil (silicided poly) |
7 Ω/□ | small series and ballast resistors |
rppd (p+ poly) |
260 Ω/□ | bias networks, feedback resistors |
rhigh (high-resistance poly) |
1360 Ω/□ | large bias resistors |
The resistors are r3_cmc devices, which Xyce’s single-plugin build cannot load: in Xyce, model them as ideal resistors of the same value.
C.9 Where each number is used
| Device | Designed into |
|---|---|
lv nMOS cascode, cells of m > 1 |
LNA (Chapter 9); passive-mixer switches (Chapter 10); VCO core (Chapter 11); CML and TSPC dividers (Chapter 13) |
| npn13G2v cascode | 2.4 GHz PA (Chapter 16) |
| npn13G2 | 50 GHz PA (Chapter 17); the HBT variant of the LNA (Chapter 9) |
| Inductors L1–L4, V4 | LNA, VCO tank, PA matching networks |
| cap_rfcmim, svaricap, switched MIM | matching networks, DC blocks, VCO tuning bank |
| TopMetal2 lines and pads | the 50 GHz PA’s matching networks, layout and EM verification (Chapter 18) |