Diamond_pcb_1st_steps

Download Report

Transcript Diamond_pcb_1st_steps

CVD PCB, first steps
Overview
Overhang:
Up to 5 mm
Chip area. No ground plane underneath the chip.
Bulk isolated => only one ground line
(if needed)
Possible location for the 100nf and 10nf
decoupling capacitors. Close to the chip
15 mm
4mm
Power lines
Bonding area
Control (bonding) pads
Connector: 11,1mm*2,1mm: Molex 502430-4410
25 mm
Possible location for the 1-10uf bulk
decoupling capacitors. Close to connector
HV
7,5 mm
Leave at least 5-10mm of clearance to be used
as contact surface for heat dissipation
4mm
Chip signals/power interface
Try to put capacitors here.
A thermal relief to
connect the capacitors
to the ground/power
lines shall be needed,
otherwise soldering can
become very difficult
due to good thermal
conductivity of silver and
diamond. Connecting
with a thin trace of
~200um) should be fine.
3mm
• The bonding map
needs to be redone due
to lack of space.
• Careful thinking about
it has to be taken.
• This point is very
important because if
needed we could even
ask for a pin-out
modification in the
future chip
Power lines dimensions:
• AVDD: 600 um
• VDD: 400 um (200 um according to PW consumption should be fine)
• DVDD: 400 um (200 um according to PW consumption should be fine)
•GND: 1.2mm should be fine. Underneath the cheap it can be a little bit wider
(1.4mm)
If possible, better 400 um width than 200 because of impedance
reasons and to allow more space for bonding space so the pcb
can be reused. Normaly 200 um is a sensible space for 1 bonding.
Vdrop calculations
 Voltage drop in power lines for silver deposition:
o Resistivity: 3 mOhm/square for 10 um thickness 
o 1 mOhm/square for 30 um thickness 
o For AVDD (~25mm length, 0.6 width)  27 mV Vdrop
o 650 mA  0.65 mV/square
o If length ~25mm and width 0.6 mm  41,7 squares  27 mV Vdrop
o For VDD (~25mm length, 0.4 width)  10.6 mV Vdrop
o For DVDD (~25mm length, 0.4 width)  3.2 mV Vdrop
o GND (~900 mA, ~25mm length, 1.4 width)  16 mV Vdrop
 The current consumption in the table is running at 200 MHz.
We’ll run at 40 MHZ!! So it will be lower. Nevertheless may have to
increase a bit the width of AVDD and/or GND.
Connector


Only 1 connector for LV, HV and signals
Connector type: Molex slimstack, 44pins, double row, 400um pitch,
max pin current 0.3A:
o
o
502430-4410 => In diamond PCB end, it is narrower (11.1 x 2.1 mm)
502426-4410 >> In FPC side (11.9 x 2.6 mm)
 Total number of pins: 44
o
o
o
o
o
o
o
o
o
o
Needs to
withstand
500 V
o
21 (11 differential signals)
3 (single-ended control signals)
1 (3.3V line)
3 (AVDD)
1 (DVDD)
1 (VDD)
4 (GND)
1 (HV bias)
2 floating (HV clearance)
2 free. To be used to connect the ground of the decoupling capacitors
Open issue. How to connect the 10uF capacitors. It doesn’t seem to be
straight forward to do it… We’ll have to struggle a bit, but just a bit.
GND
Floating
HV bias and filter:
Place a pad somewhere in the HV path so we could solder a wire in case we wanted
bias the sensor externally. A 100K resistor and a 22nf càpacitor able to withstand
500V needs to be placed here. It may be difficult to find de capacitor…
Heat dissipation
Top view:
Side view:
Connector
Diamond PCB
At least 5mm
Cooling metallic support (Alu?)
Technical considerations
• 200um track and gap is the 'standard' that can
easily be achieved.
• 100 um gap/width is possible (so connecting to a
0.3mm pitch connector is feasible). Nevertheless,
let’s try not to push to the limit.
• No vias => one-side board without through-holes.
• The chip might be powered only from one side.
• The metal used is silver. It can get oxidized.
Maybe we’ll have to think about a coating…
• The square resistance is ~3 milli-ohm/square.
Ideas
• No ground plane beneath the chip since the
bulk of the device is isolated. Besides, silver
can get very activated.
• The 100nF and 10nF must be close to the chip,
being the 10nF the closest. They have to be
ceramic and the smallest possible (0402
should be fine). The bulk capacitors of 1 to
10uF must be close to the connector. If 10uF
do not exist in 0402 size we could use lower
capacitance values. They must also be
ceramic.
Ideas
• We can get rid of most of the DOUT
differential pairs as we don’t read-out in
parallel mode yet. According to the medipix
manual only dataout[0] output is needed. This
reduces the number of differential pairs to 11.
• It would be nice to have a clearance area of 510 mm at least in one side of the diamond to
use it as thermal conductive contact to a
clamp.
• We’ll have to check the performance of
thermoplastic glue (or whatever is used) in
diamond…
Very 1st VERSION
Remove ground plane.
Bonding impossible if routed this way. Bonds too long and
capacitors are an obstacle.
Put termination resistors in zig-zag ?
Differential lines should be equally spaced
Use the mating connector, is smaller
• Very first version. Many things to change.
• We’ve learnt so far:
• We lack of space.
• Use only one connector
• The layout can/must be improved.
• Let’s see if we manage to route
everything in one layer