For the purpose of saving RF pads and for the convenience of measurement, LO driver and Differential to Single (D2S) circuits were designed and fabricated on chip together with the mixer. Besides, these auxiliary circuits are necessary for a realistic optical front-end system which intends to integrate more function
blocks on chip. The LO driver schematic is shown in Fig. 4.17. M1 and M2
form the differential pair to amplify the input LO from off-chip. The gate of M2
is AC grounded by a 7 pF capacitor. Resistor Rb provides DC operating point
of M2 by shorting the gates of M1 and M2. In this design, Rb is implemented
by poly2 resistor and has a value of around 4 KΩ which together with the AC grounded capacitor form a low pass filter with a cut-off frequency of 6 MHz. For measurement set up of the mixer, DC bias voltage of the LO driver has been
chosen to be around 1.4 V. It guarantees a reasonable Vds for M1 and M2, at a
tail current of 5 mA (2.5 mA for each branch of the differential pair), which is
resistors, for the consideration of the gain.
Figure 4.17: Schematic of the LO driver
To achieve the maximum gain for the circuit, an inductive load would have been preferable for the LO driver since it give better impedance and gain at high frequencies, but this comes with the cost of large chip area, which is unbearable because of the unavoidable inductors in TIA and input matching circuits of the mixer. Besides, another inductor is reserved for the LO driver which provides matching from off-chip to the input of the circuit. Simulation shows that at 5 GHz, appropriate resistor load still provides gain for the LO driver. With a tail current of 5 mA, the load resistor is chosen to be 300 Ω which is optimized for bandwidth, gain and the DC coupling to the LO ports of the Gilbert mixer.
It is difficult to have the two outputs of the LO driver with the same amplitude because of unmatched input to the differential pair, and the gate to drain parasitic capacitances Cgd of M1 and M2. In fact, Vout p has a larger amplitude than Voutn.
Fortunately it does not impose a big problem so long as the output amplitude is large enough and the two outputs are out of phase. The design values for the components are listed in table 4.1.
component design parameter design value M1 30× 5/0.35µm M2 30× 5/0.35µm Ms 30× 5/0.35µm R 6× 100/3µm 1818Ω/6 Rb 2 *120/3µm 2182Ω×2 L 4.7 nH 4.7 nH Cac grounded 7× 33.9 × 33.9µm 7 pF
Table 4.1: Design values of the LO driver
The AC simulation results of the LO driver are shown in Fig. 4.18. Without inductive peaking, the LO driver demonstrates 7.7 dB voltage gain at 5 GHz, which greatly reduces the power requirement of the LO input.
Both the prototype chips of the mixer subcircuit and the optical front-end were not packaged for this project, and all the measurements were to be conducted with probe station. G-S-G probes and needles probe are used for RF signal coupling (both input and output) and DC connection, respectively. The probes are expected to be as few as possible since their mechanical arms occupy a lot of space on
the test platform4and the system would get congested. Therefore the differential
output from the proposed Gilbert cell is preferred to be converted to single ended one. This was done by the Differential to Single (D2S) circuit, as shown in Fig. 4.19. It is basically a single stage amplifier followed by an output buffer. Since the signal has been downconverted to IF by the mixer, the design constraint on the
bandwidth of D2S has been relaxed. The tail current for the differential pair M1
and M2in Fig. 4.19 can be chosen smaller than that in the LO driver, 3 mA for this
design. M5and M6 form source follower which provide 50 Ω impedance match
to the measurement instruments. Note that the source follower output impedance can be roughly written as 1/gm, to have more gain at the output resistor RL, it is
expected gmto be chosen as large as possible because the source follower gives a
gain of gmRL/(1 + gmRL). On the other hand, to avoid distortion at the output, DC
operating point should be close to the mid of power supply, which results a large DC current through M5, M6and RL.
For the mixer in RF15, however, some design issues in the D2S circuit were
discovered after the tape out. First, the load resistor R can be replaced by PMOS transistors which saves chip area; secondly, for the D2S output in Fig. 4.19 (the
load RLis 50 Ω to emulate the input impedance of the measurement instruments, i.
e., spectrum analyser), it must be DC coupled to the spectrum analyser, otherwise,
there is no DC current supply for M5and M6. But most of the spectrum analyzers
have DC block at the RF input port for DC protection. In the tape out of the optical front end (tape out series No. Atto1b), these issues were removed and the D2S circuit was re-designed.
4The probe station used for the measurement is not a commercial one, instead, it was built up
manually in the EEE Department, University of Nottingham.
5RF1 is the MPW (Multi-Project Wafer) tape out series number, which includes prototype
circuits for RF applications, i.e., 5 GHz frequency divider for frequency synthesizer; 4 GHz LNA for breast cancer imaging system, and 5 GHz mixer for the high speed optical front end.
Figure 4.19: Schematic of the Differential to Single circuit
component design parameter Design value
M1,2 20× 5/0.35µm
M5,6 40× 5/0.35µm
MID 12× 100/3µm
R 4× 120/3µm 2182 Ω /4
Table 4.2: Design values of the LO driver
The simulation result of D2S is shown in Fig. 4.20. As the IF frequency is around 100 MHz, D2S provides about 6.9 dB gain. These should be deducted from the mixer measurement results to get the net gain of the mixer core. The simulation results did not take into account of the effects of parasitic capacitances, i.e., that from the pads and wirings. Its - 3 dB bandwidth is estimated to be in fact about few hundred MHz, which means that if higher IF frequency is chosen, more current needs to be assigned to the D2S circuit to improve its bandwidth.
Figure 4.20: AC simulation result of the D2S