A PRACTICAL APPROACH IN MODELLING SILICON-CROSSTALK IN SYSTEMS-ON-SILICON. PAUL T.M. VAN ZEIJL ERICSSON EMMEN THE NETHERLANDS
[email protected]
S eptember 6, 2001
1
Table of contents. • Introduction: problem statement. • Approaches to Si-crosstalk: state-of-the-art and its problems. • Our strategy. – – – –
Modelling digital circuitry. Modelling pure CMOS substrates. Modelling BiCMOS/RFCMOS substrates. Modelling analog circuitry.
• Requirements on digital and analog design. • Conclusions.
S eptember 6, 2001
2
© E rics s on, Paul T .M. van Z eijl
Problem Statement 1(3). • Possible causes of crosstalk: – – – –
Crosstalk on the PCB. Crosstalk via the bonding wires and package. Crosstalk via ground and supply lines. Crosstalk due to the Si-substrate.
• Consequences of interfering signals due to crosstalk: – Addition of signals (linear). – Modulation of signals (non-linear): • May create extra spurious components. • May shift bias points. • May give pushing/pulling effects on VCO/XO.
• We don’t have ways to easily predict the magnitude effect(s) of crosstalk-on-silicon. S eptember 6, 2001
3
© E rics s on, Paul T .M. van Z eijl
Problem Statement 2(3).
S eptember 6, 2001
4
© E rics s on, Paul T .M. van Z eijl
Problem Statement 3(3). • Options for minimising Si-crosstalk: – Separate in the frequency domain (this may prove difficult due to high data-rates). – Separate in the time domain (no digital activity during reception/transmission of signals). – Lower amplitude of interference source (introduce jitter on clocks). – Isolate (use extra layout measures such as shielding or triplewell). – Compensate or balance.
• BUT how much improvement do we get from these measures? S eptember 6, 2001
5
© E rics s on, Paul T .M. van Z eijl
Approaches to Si-crosstalk: state-of-the-art limitations • Large number of digital gates (> 1M); not practical in analog simulations. • Fast slopes in digital circuits in state-of-the-art CMOS processes. • Complexity in substrate extraction (practical size and simulation time for the substrate model). • Low-frequency digital (~ MHz) vs analog/RF (~ GHz). • Can only done after layout has been finalised. • Usually done after receiving 1st silicon, because there are “problems”. • For a lot of designers it is like “black-magic”. S eptember 6, 2001
6
© E rics s on, Paul T .M. van Z eijl
Our strategy. • Start during the design phase of the ASIC. • Use the knowledge of the digital circuitry/functionality to start modelling it; use a simple model. • Use the floorplan to get a model for the substrate. • Specify analog/RF circuitry properly. • Use the analog circuitry and simulate for interference sensitivity. • Use a simple model for the digital circuitry, the model for the substrate and the analog circuitry to do analog simulations on the overall performance. Compare with specifications. S eptember 6, 2001
7
© E rics s on, Paul T .M. van Z eijl
Example of specification of VCO pulling.
∆f = 5 kHz when switching RX/TX mixers ON Ref.: P.T.M. van Zeijl et all, ESSCIRC, Southampton, 1997 S eptember 6, 2001
8
© E rics s on, Paul T .M. van Z eijl
Modelling digital circuitry 1(4). • Model the digital interfering source by: – Icc at given Vcc. – Frequency. – Is the behaviour dominated by clocked signals or more like a pseudo-random-bit-sequence (PRBS)? – Use simple LARGE inverter to speed up simulation. – If needed model VCC/GND capacitive coupling to substrate.
S eptember 6, 2001
9
© E rics s on, Paul T .M. van Z eijl
Modelling digital circuitry; clocked signals 2(4). Spectrum of clock signals: f_clock=10 MHz t_rise=t_fall=1ns duty-cycle=45/55. f_clock= 10MHz, t_rise=t_fall=100ps duty-cycle=45/55. Frequency from DC to 2.5 GHz. Amplitude from -100 to 0 dB. S eptember 6, 2001
10
© E rics s on, Paul T .M. van Z eijl
Modelling digital circuitry; PRBS 3(4). Spectrum of PRBS signals: f_clock=10 MHz. t_rise=t_fall=100p. Frequency from DC to 2.5 GHz. Amplitude from -100 to 0 dB.
Frequency from DC to 50 MHz. Amplitude from -50 to 0 dB.
S eptember 6, 2001
11
© E rics s on, Paul T .M. van Z eijl
Modelling digital circuitry; I_substrate 4(4). 1k inverter: • Substrate voltage, 1 nH series inductance to PCB-gnd, 10 MHz clock, 1 Vpp. • Substrate short-circuit current, +20 mAp, -40 mAp.
• Spectrum of substrate voltage: resonance at 4 GHz: 1 nH//1.3 pF. Frequency from DC to 5 GHz. Amplitude from -100 to 0 dB.
S eptember 6, 2001
12
© E rics s on, Paul T .M. van Z eijl
Substrate modelling 1(8); low-impedance substrate: 0.35u pure CMOS. • Contents of ASIC: – Analog blocks: LNA, VCO, front-end, etc. – Digital blocks: large inverters, at various places in the layout to check distance dependency.
• Inverters are driven by sinusoidal signals so that crosstalk on the PCB, or crosstalk due to the bonding wires and package is avoided.
S eptember 6, 2001
13
© E rics s on, Paul T .M. van Z eijl
Substrate modelling 2(8): 0.35um pure CMOS; low-impedance substrate.
1998 S eptember 6, 2001
14
© E rics s on, Paul T .M. van Z eijl
Substrate modelling 3(8): simulation model for 0.35u pure CMOS substrate.
S eptember 6, 2001
15
© E rics s on, Paul T .M. van Z eijl
Substrate modelling 4(8): comparison of measurements versus simulations.
T hird harmonic: Meas ured: * S imulated: o
0.35um pure CMOS
F requency from 1 MHz to 2 GHz. Amplitude from -50 to =100 dB uV. S eptember 6, 2001
16
© E rics s on, Paul T .M. van Z eijl
Substrate modelling 5(8): 0.35u pure CMOS substrate; conclusions. • A pure CMOS substrate can be modelled as a short. • Impedance to PCB-ground (bonding-wire and package) should be taken into account (including series resistance and inductance). • Separation distance is not important.
S eptember 6, 2001
17
© E rics s on, Paul T .M. van Z eijl
Modelling BiCMOS/RFCMOS substrates 6(8).
• Buried-N, triple-well can be used for extra isolation. • Sometimes extra mask, sometimes part of standard processing. • Modelling substrate: – Estimate area per block (from floorplanning). – Estimate number of substrate contacts per block. – Use tools to generate netlist model.
S eptember 6, 2001
18
© E rics s on, Paul T .M. van Z eijl
Modelling BiCMOS/RFCMOS substrates: example 7(8).
Acknowledgement goes to: Didier Belot from STMicroelectronics Francois Clement from Simplex
S eptember 6, 2001
19
© E rics s on, Paul T .M. van Z eijl
Modelling BiCMOS/RFCMOS substrates 8(8): conclusions. • Remarks when generating such a substrate model: – Minimise the number of substrate contacts. – Up to ~ GHz, the substrate model can be considered purely resistive. – Minimise the size of the netlist (32 Mbyte vs 1 Mbyte). – Limit the values of the elements in this netlist to reasonable values and range.
S eptember 6, 2001
20
© E rics s on, Paul T .M. van Z eijl
Modelling analog circuitry.
S eptember 6, 2001
21
© E rics s on, Paul T .M. van Z eijl
Example of simulation testbench.
S eptember 6, 2001
22
© E rics s on, Paul T .M. van Z eijl
Requirements on digital and analog circuitry.
• • • • • • •
Specify PSRR for all analog/RF circuits. Specify CM-rejection for all analog/RF circuits. Specify CM⇒DM conversion for all analog/RF circuits. Specify pushing/pulling for oscillators and VCO’s. Separate power supplies. Use shielded bonding-pads or shielded bump-pads. Use bumping for very low impedance to PCB-ground.
S eptember 6, 2001
23
© E rics s on, Paul T .M. van Z eijl
Conclusions.
• • • •
Do pre-layout simulations Create problem awareness in design-team Get input from digital designers. Si-crosstalk has to be taken into account from the start: one more issue on the list of the analog designers, increasing design complexity. • Consequently: more compromises in (RF) design! RF design becomes (even more) multidimensional.
S eptember 6, 2001
24
© E rics s on, Paul T .M. van Z eijl