TWEPP-2009, Paris, September 2009
Download
Report
Transcript TWEPP-2009, Paris, September 2009
[email protected]
TWEPP-2009, Paris, September 2009
Ultra Thin, Fully Depleted MAPS based on 3D Integration
of Heterogeneous CMOS Layers
Wojciech Dulinski on behalf of:
IPHC Strasbourg-IRFU Saclay-University of Pavia-University of Bergamo
Collaboration
On the way towards fast, high precision, radiation tolerant and ultra thin
CMOS sensors, we propose MAPS on fully depleted epitaxial substrate with
first stage buffer amplifier on the same wafer, capacitively coupled to the 3D
readout electronics implemented on top of each pixel.
Fast: frame readout time <<10µs and/or time resolution of ~100 ns
High precision: pixel pitch <20 µm, spatial resolution ~2µm
Radiation tolerant: > 1014 n/cm2
Ultra thin: ~50 µm (Si equivalent)
1
[email protected]
TWEPP-2009, Paris, September 2009
Recipe for our first, on-going exercise:
use Tezzaron/Chartered 0.13 µm 2-tiers process available through
the HEP-3D Consortium (Fermilab, IN2P3, INFN…)
plus XFAB PIN (high-resistive epi) process
Bonding pads
Bumps
TSV
From Chartered
(2 tiers)
plus XFAB and Ziptronix
(3 tiers)
Wafer view at intermediate and after final stage
(chip-to-XFAB wafer bonding)
2
[email protected]
TWEPP-2009, Paris, September 2009
Existing “3D integration” technology: hybrid sensors
Bump-bonding (indium or solder bumps): minimum pitch
~few tens of microns, limited to integration of two layers,
not really industrial (large volume) process
Example: ATLAS pixel sensors
3
[email protected]
TWEPP-2009, Paris, September 2009
On-going evolution and improvement of bump-bonding:
example of micro-bonding from IMEC
(credit to Piet De Moor)
ZyCube Co. Ltd.
(Japan): 5 µm pitch
Major disadvantage of bump-bonding persists, plus a new complication
(intermetallic compounds formation: variable reliability)
4
[email protected]
TWEPP-2009, Paris, September 2009
Our choice: industrial (or close to) 3-D integration process,
with a standard use of TSV (through-silicon-vias) and
direct metal-metal thermo compression bonding
Example from Tezzaron*: two or more stacked wafers (tiers). Typical
interconnection pitch < 5 µm, typical TSV Cp < 5 fF, typical TSV Rs < 0.5 Ω,
typical interconnecting metal fill factor ~30%.
* B.Patti, Proc.IEEE Vol.94, No. 6, June 2006
5
[email protected]
TWEPP-2009, Paris, September 2009
3-D integration technology is offered by Tezzaron/Chartered
as a standard technology (similar constrains as for VLSI
submission), but limited (at least for us as a small user) to
one CMOS process (0.13 microns)
To bond our wafers from another CMOS supplier (XFAB
OPTO PIN, well suited for the sensor layer) we plan to use
another 3D-integration semi-industrial process:
DBI® (Direct Bond Interconnect) from Ziptronix
6
[email protected]
TWEPP-2009, Paris, September 2009
DBI® (Direct Bond Interconnect): Low Temperature
CMOS Compatible Direct Oxide Bonding For Highest
Density 3D Interconnection (<1 µm pitch possible)
- very low mass (>95% of bond is silicon oxide)
- mechanically stronger than silicon allowing bonding of thinnest
possible CMOS layer (~10 microns)
- directly compatible with any modern CMOS BEOL process
having W-plugs and CMP planarization steps (but also compatible
with Copper Interconnect)
Example of 3 µm diameter DBI®
( metal plug via through the oxide)
* P. Enquist, Proc. FEE-2009)
7
[email protected]
TWEPP-2009, Paris, September 2009
DBI® Process flow (from Paul Enquist talk at FEE-2009)
8
[email protected]
TWEPP-2009, Paris, September 2009
Second stage: thinning to <50 µm
Bonding pads
XFAB and 2xChartered
(3 tiers, first stage)
Problem: how to handle, interconnect and at the end built a
ladder with such a thin device?
9
[email protected]
TWEPP-2009, Paris, September 2009
Possible solution: embedding in thin flexible substrate
(BCB, silicone, polyamide). Interconnect at bondpad level
using electroplated Cu. No wire bonding!
Fully functional microprocessor chip in flexible plastic envelope.
*Curtsey of Piet De Moor, IMEC, Belgium
10
[email protected]
TWEPP-2009, Paris, September 2009
Embedding in flexible substrate process developed end
proposed by IMEC. For later this year, we are planning the
first exercise using standard, analog readout MAPS
(Mimosa18) thinned down to <30 µm, embedded in plastic
and interconnected to PCB using flex cable. The goal is to
study performance of a “flexible” sensor…
Possible solution for a cylindrical vertex detector layer?
Piet De Moor, FEE-2009 Proceedings
11
[email protected]
TWEPP-2009, Paris, September 2009
1. Self Triggering Pixel Strip-like Tracker (STriPSeT)
Collaboration: Strasbourg-Bergamo-Pavia
If
>1 nA
G
x1
Cf
Cc
>100 fF
~10 fF
soff
<10 mV
Dff
~ 40 µm2
Low offset,
continuous
discriminator
Readout logic
Cd
~ 10 fF
Qmin
~ 200 el
Shaperless
front-end
tpeak ~1µs
XFAB 0.6µm PIN (Tier_0)
Chartered Tier_1 (analog) and
Ziptronix
Tier_2 (digital)
Tezzaron
(Direct Bond Interconnect, DBI®)
(metal-metal (Cu) thermo compression)
12
[email protected]
TWEPP-2009, Paris, September 2009
Principal arguments for use of XFAB-0.6 PIN
- “fully” depleted, 14 µm thick epitaxy
- for small pitch, charge fully contained in less than four pixels
- fast charge collection (~5ns) should be radiation tolerant
- sufficient (rather good) S/N ratio defined by the first stage
- “charge amplification” ( >x10) by capacitive coupling to the second stage
- first experimental results from our first pixel sensor in this process (Mimosa25)
very promising!
20x20 µm pixel layout in XFAB-06.
Version: SF+CAPA (150 fF)
13
[email protected]
TWEPP-2009, Paris, September 2009
TCAD simulation of a MAPS on high resistivity (1 kΩ cm) epi
For comparison: standard
CMOS technology, low
resistivity P-epi
high resistivity P-epi: size of
depletion zone size is
comparable to the P-epi
thickness!
14
[email protected]
TWEPP-2009, Paris, September 2009
Mimosa-25 prototypes (XFAB-0.6 PIN, Aug-Dec 2008)
20um
4x4 STD
30um
11.4x11.4
PIN
20um
4x4 um
STD_TOX
40um
11.4x11.4 um
PIN
20um
5x6.5 SB
30um
5x6.5 SB
20um
5x6.5 um
SB_TOX
40um
11.4x11.4um
STD
20um
5x6.5 STD
30um
4x4 STD
20um
5x6.5 um
STD_TOX
40um
5x6.5 um
STD
Mimosa 25 A
Mimosa 25 B
2 submitted chips
15
[email protected]
TWEPP-2009, Paris, September 2009
Mimosa-25 on fully depleted epi substrate (XFAB): first tests results (Ru beta)
20 µm pitch, self-bias [email protected] before and after neutron irradiation
Landau MP charge of the cluster versus cluster
size before and after neutron irradiation
To compare: « standard »
(non-depleted epi substrate)
before and after ~6*1012 n/cm2
16
[email protected]
TWEPP-2009, Paris, September 2009
Mimosa-25 on fully depleted epi substrate : Ru beta spectrum (MIP Landau)
seen at the seed pixel, 20 µm pitch, self-bias [email protected]
Supposed threshold for ~100% efficiency: ~150 electrons
very recently confirmed by our beam tests!
17
[email protected]
TWEPP-2009, Paris, September 2009
Principal arguments for “shaperless front-end” (single
stage, high gain, folded cascode based charge amplifier, with
a current source in the feedback loop)
- simple (surface efficient) but very satisfactory approach, in particular when
minimum signal charge is of few thousand electrons (after “charge
amplification”)
- shaping time of ~1 µs very convenient: good time resolution, insensitivity to the
irradiation induced leakage current
- possible implementation of Time-over-Threshold ADC in the future…
- minimum signal at the entrance of comparator few tens of mV, so threshold
dispersions of few mV tolerable
- structure already studied in details by Pavia&Bergamo group (expertise from
several prototypes in another 0.13 µm process): minimum risk for this first
3D-3T exercise!
18
[email protected]
TWEPP-2009, Paris, September 2009
Combined simulation XFAB+Charterer (SF+ShaperlessFE)
Two versions of SFE: Cf=11 fF (MiM) and 5.5 fF (VPP)
Simulation:
pulse corresponding to 200 el
injected into XFAB diode
Results:
Peaking time ~400 ns
Gain: ~150µV/el (300 µV/el)
ENC ~10 electrons!
Extra power from SF:
<1µW/pixel
IF this is correct,
we should expect:
S/NLandau>40
and
Cut~100%eff/Noise>10
19
[email protected]
TWEPP-2009, Paris, September 2009
Combined simulation XFAB+Charterer (SF+ShaperlessFE)
Linearity: 100-4000 electrons (step 100 e)
20
[email protected]
TWEPP-2009, Paris, September 2009
Combined simulation XFAB+Charterer (SF+ShaperlessFE)
Comparator output : 100-4000 electrons (step 100 e); threshold ~150 e
21
[email protected]
TWEPP-2009, Paris, September 2009
Combined simulation XFAB+Charterer (SF+ShaperlessFE)
Time-walk of the comparator output : 100-4000 electrons (step 100 e); thr. ~150 e
22
[email protected]
TWEPP-2009, Paris, September 2009
Self Triggering Pixel Strip-like Tracker: analog pixel (SFE) layout
SuperVias
(Input and GND)
Input
23
[email protected]
TWEPP-2009, Paris, September 2009
Principal arguments for “only digital” Tier_2
- excellent separation of analog and digital (no common substrate, several metal
layers for blinding…): no problems for asynchronous, random logic
- flexibility of the readout architecture, possible use of “front-line”, pure digital
CMOS process (<60nm) for this layer to increase complexity of processing at
lower power budget
24
[email protected]
TWEPP-2009, Paris, September 2009
STriPSeT: Data driven (self-triggering), sparsified binary readout.
X and Y projection of hit pixels pattern
SR
TrigOut
Example:
DELAY
SR: hot pixel disable register
TrigIn
readout clock : 160 MHz
2 output lines ≡ Array readout
SR Readout or Reset time: ~2µs
Programmable Active Area
(through pixel disable SR)
Readout compatible with existing IPHC-digital DAQ…
25
[email protected]
TWEPP-2009, Paris, September 2009
2. Rolling Shutter Mode MAPS: power efficient solution
based on M26 approach for processing.
Collaboration: IRFU (Saclay) –IPHC (Strasbourg)
In-pixel electronics
- Common threshold voltage
- Only NMOS transistors in Tier 1
- 20x20 µm pitch, 32x256 pixel array
- Low power operation (rolling shutter)
Example of a power budget:
-100 µW/pixel for 50ns processing
-To be compared with 500µW/pixel for 200 ns processing (Mimosa26)
-Factor of 20 saving in analog power! This is due to 3D electronics…
26
[email protected]
TWEPP-2009, Paris, September 2009
3-Tiers CMOS MAPS status and expected (revised) schedule
(May today)
Submission (Chartered): ~April (done) mask generation at Chartered just now (?)
-
Submission (XFAB): ~June/July (building blocks ready) changing to XFAB 035 PIN
process, expected submission before end 2009
-
-
Reception of wafers: ~ September December/January 2009
-
2 Tiers testing: ~September from January 2100 on
XFAB Tier bonding (at Ziptronix): October revision of technology compatibility
-
Final tests (including beam tests): before the end of 2009? second half 2010?
Next steps:
1.
2.
Power optimization. Power dissipation is a major issue in case of very thin detectors!
In addition to STRipSeT and RSPix, a novel, high-gain (1-2 mV/el), low power
(<1µW/pixel) test structure has been submitted
Thinning down to ~50 µm. Having such thin wafer, we may propose a new approach to
solve yield problems in case of large area devices: built-in redundancy (2complete layers)
plus powering off of selected areas after in-situ self-tests
27