HAL Id: jpa-00249394
https://hal.archives-ouvertes.fr/jpa-00249394
Submitted on 1 Jan 1995
HAL is a multi-disciplinary open access archive for the deposit and dissemination of sci- entific research documents, whether they are pub- lished or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.
Interconnections on Prediffused ASIC’s
Agnès Fleury, E. Saint-Christophe, H. Frémont, M. Fathi, G. N’Kaoua, Y.
Danto
To cite this version:
Agnès Fleury, E. Saint-Christophe, H. Frémont, M. Fathi, G. N’Kaoua, et al.. Laser Beam Lithogra- phy for Direct Patterning of Interconnections on Prediffused ASIC’s. Journal de Physique III, EDP Sciences, 1995, 5 (9), pp.1455-1467. �10.1051/jp3:1995203�. �jpa-00249394�
Classification Physics Abstracts
43.20M 05.60
Laser Beam Lithography for Direct Patterning of Interconnections on Predilllused ASIC'S
A. Fleury, E. Saint-Christophe, H. Frdmont, M. Fathi, G. N'Kaoua and Y. Dante Laboratoire IXL, 351 cours de La Lib4ration, 33405 Talence Cedex, France
(Received 27 December1994, revised 15 March 1995, accepted 19 June 1995)
Abstract. A lithography machine designed for laser direct writing of metallisation in ASIC'S and MCM, and the associated practical process are described. To demonstrate the feasibility
of the laser direct writing, the last level of metallisation on
a bipolar ASIC has been drawn.
The design rules have been calculated using a theoretical modelling of the interaction between laser beam and photoresists. Calculation and experiment are in good agreement for two kinds of photoresists.
1. Introduction
The final step of the fabrication process of a prediifused ASIC, I-e- the customisation, is the addition of one, two, or sometimes three levels of metallisation. In the same way, the interconnection of dies on a thin film hybrid Multi-Chip-Module necessitates the design of several levels of metallisation. Direct writing of these interconnections during the development phase is cheaper than using a classical optical mask and permits to adjust the layout before
producing an expensive, but optimised final optical mask. Our objective was to develop new prototypes or test structures using a laser beam direct writing technique. The spatial resolution of the writing (Wr) must fit the dimensions of LSI or VLSI circuits, whose connections are a
few micrometers wide. Hence the visible laser beam is sufficient (0.8 ~Lm < Wr < a few ~Lm)
although the writing resolutions are finer with an excimer laser beam (0.2 ~Lm < Wr < I ~Lm)
or an electron beam (3 nm < Wr < 0.2 ~Lm) [1-3]. So, a Hecd laser beam, of wavelength
~ = 442 nm (visible range, blue), was taken because of the following reasons:
the etch resolution matches the micrometric range, this kind of laser is cheaper than an electron beam,
no vacuum is needed.
Supported by a practical study, whose purpose is to demonstrate the feasibility of the di- rect writing by doing the last level of metallisation on a bipolar ASIC, this paper presents a
theoretical calculation of the width of the final metal lines, based upon a modellisation of the interaction between the laser beam and the photoresist.
@ Les Editions de Physique 1995
Videocamm
mm
pz SUN
ETHERNET
Fig. 1. Schematic diagram of the laser lithography bench. The setpoint for the power of the Hecd laser in adjusted using the aperture angle of the polariser.
2. Experimental
2. I. APPARATUS. Figure I shows the full schematic diagram of the laser lithography bench [4]. The main part is a Hecd laser, having an output power of16 mW, leading to an effective value Po of1.5 mW on the sample. The optical microscope works as a filtering and focusing system and allows a spot of about 0.9 pm to be focussed on the wafer. The laser,power P at the 1nlpact point can be adjusted using a polarizer located on the optical path of tie beam.
The relationship between P and the aperture angle of the polarizer is:
P = Po cos~@. (1)
varies from 0 to 90°. During the patterning sequence, the beam can be split using an acousto- optic modulator. A diode laser of wavelength ~
= 780 nm is used for focussing, thus avoiding
defaults due to the wafer topography. A micropositionner is used to control the movement of the wafer in the z, y and z directions. It has a resolution of 0.I ~Lm, and a programmable velocity between 10 pm s~~ and 25 pm s~~ The
z axis is used to focus the laser beam. The whole bench is controlled by a computer and an external Work Station is used to design the
masks (EDGE software) that are transfered to the computer via a network.
The CAD patterns have to be physically reproduced on the substrate taking into account the preceeding layers in order to perform a correct alignment. The techniques based on a
common stamp for the CAD design and the substrate, need stamped substrates, that is to
say an additional step for standard substrates [5]. Methods using a TV type scanning are based on the coupling between a mechanical move of the substrate and shifts of the laser beam [6, 7]. They are faster than the previous method but need a complex software treatment
a) b)
Oncwanforoa&of~csc 2wacdQV=D)
Two mm for I
j rifts ~ack
0V=2D)
fictive
14
fi 02
Fig. 2. Decomposition of a simple pattern Figure 2a into elementary rectangles (Fig. 2b). Each
one is identified by W~, and the location of 0i. Scans follow l~. D is the efficient beam diameter.
because of a pixel writing. The system described in this paper uses a vectorial writing mode.
The CAD patterns are considered as a set of tracks and the basic elements are rectangles (EDGE software). The software that drives the whole bench is written in C language. The pattern shown in Figure 2a is for instance decomposed by the software into 3 basic rectangles
as depicted in Figure 2b. Each rectangle is identified by its origin 0~ relatively to the point
0 (bottom left of the pattern) used as alignment sight, and by its length and its width W.
The direct writing scans to and fro each element; the number of scannings depends on the width of each rectangle and of the laser beam efficient diameter D: for instance, in Figure 2b, polygon 03 is scanned with two line scans and the other ones only one time. The accuracy of positioning is assured by the resolution of the shift axes, that is to say 0.I pm. The precise starting point is assured by the fact that the laser beam is switched on just as soon as the
velocity of the axes is stabilized, at the boundary of the rectangle (to reach this velocity, it takes about 1.25 ms). The successive patterning of every layer is performed by linking the software to a manual movement of the axes (for alignment purpose) and an automatic drawing
of the CAD pattern.
2.2. PHOTORESISTS AND ASSOCIATED PRocEssEs. In the case of a negative photoresist,
a conductive layer is first deposited on the substrate and covered by a photoresist layer. After exposure and development of the resist, the conductive layer is etched, either wet (chemistry)
or dry (plasma etching). The desired interconnections remain (Fig. 3a). This process seems to be the simplest. However, existing negative photoresists may be dangerous products and for
that reason difficult to use. Moreover, they are not sensitive to a wavelength of 442 nm.
A positive photoresist could be used in the same way, but the whole wafer must be exposed, except for the areas to be kept. But as areas without any interconnection are much larger than
those covered by the interconnections, the insolating time becomes prohibitive.
Another process using a positive photoresist, called lift-off, is depicted in Figure 3b: the conductive layer is deposited on the substrate already covered by the exposed and developed photoresist. If the conductive layer is thinner than the photoresist layer, wet etching of the
photoresist is possible, and the upper layer is removed at the same time. The major restriction of the later process is due to the equivalent thicknesses of the two layers. In most cases, the
thickness of the metal lines must be in the 2, 5 pm range, in order to lower the line resistance.
C) ~'~**
-v~-- .--~~
~~ --- .---
~- --
iii "'''
a) $j~i :iiiflile~z
f
ll~.I~
.Re~~~ev+~-
~
.xmw-~m
fi
-Mew-;-
~
-Retwtadpp~g
~
m~÷tsw
~
@ -OWW~_ subs÷ts÷e
""''"~'
"""'" ~V~"Ye' g""'&
#~~~
~ '~'~PhW°*'Is÷ ~it~~÷Q~$~~~
Fig. 3. a) Metal patterning using negative photoresist; b) Metal patterning using a lift-off process;
c) Metal patterning using an image reversal process.
Unfortunately thicknesses of a few micrometers of photoresist are very difficult to bake and expose correctly.
This is the reason why we used a fourth technique, called the inversion technique. A basically positive photoresist is used but it is transformed into a negative one by addition of1.5i~
Imidazol [8]. A UV flood exposure is necessary after the laser exposure and the post-exposure baking to perform this inversion. The other steps of the process are the same as those used
with a negative photoresist (Fig. 3a), as shown in Figure 3c.
2.3. APPLICATION
2.3.1. Definition of the Design Rules on Planar Wafers. In order to set-up the design rules
on the system, the characterization of the bench was first realised on a 2" silicon wafer covered with I pm of sputtered aluminium.
The required line width can be obtained using various methods. For a constant scan speed,
some examples are:
one scan of the focused laser beam, with a low aperture angle @i of the polariser corre- sponding to a "high" laser power,
parallel scan of the focused laser beam with an aperture angle @2 of the polariser corre- sponding to a lower laser power,
one scan of the defocused laser beam, with an appropriate setpoint of the aperture angle.
If the required line width W is small enough to be obtained in one scan, depending on the laser power Po and on the photoresist used, the first method will be the simplest one. The
A
No ovedap bePveen B
A and B
&)
»
b)
Fig.
Fig. 4. Two scans with the focused laser beam (D was 4.5 pm for one scan). Results af-
ter Alu etching. Open circuits appear. (experimental conditions: Photoresist 1400-27; 9
= 86°,
P = 7.3 x 10~~ mW; development: 5 ruin in microposit developer of shipley diluted with water 50$l in volume; Al etching time determined visually).
Fig. 5. Design of a right angle either without overlap (a) or with an overlap (b).
last two solutions are interesting for large line widths. If the second method is used, must be set so that D is greater than the CAD designed line width W divided by the number of
scans. For instance, in the application where a 9 pm line width is required, D must be greater
than 4.5 pm. If no overlapping is performed, open circuits may appear after the etch step (Fig. 4) due to the fact that the edges are not perfectly regular. The defocusing of the beam
(third method) must be adjusted experimentally and is for this reason not easy theoretically
to predict and hence to implement.
It is also necessary to define specific CAD rules concerning the overlap of horizontal and vertical lines at their junction. In fact, a pattern including a right angle is considered by
the software EDGE as two rectangles A and B (Fig. 5). They either may not overlap at all
(Fig. 5a) or totally overlap (Fig. 5b). In practice, the first solution involves a bad connection and a risk of open circuit during the life of the component (Fig. 6). The total overlap solution
provides good results as shown in Figure 7. To avoid irregularities on the pads bondaries, they
may be designed as a square boarded with 4 elementary rectangles, as shown in Figure 8.
2.3.2. Specific Problems and Solutions on ASIC'S Results. Tests performed on real in-
tegrated circuits instead of bare wafers introduce a new difficulty: the photoresist spreads laterally on some parts of the wafer above areas with inclined relief, which causes short-circuits
between parallel tracks after metal etching (Fig. 9). However, this problem does not occur if these areas are crossed over. This behaviour is explained in Figure 10: the laser beam is
Fig. 6 Fig. 7
Fig. 6. Alu etching without overlap. Risks of open circuits appear. (experimental conditions:
Photoresist 1400-27; 9
= 55°, P = 0.49 mW but defocussed laser beam; development: 5 min in
microposit developer of Shipley diluted with water 50$l in volume; Al etching time is determined visually).
Fig. 7. Alu etching with overlap. The right angles are correctly connected. (same experimental conditions as those taken for Fig. 6).
~~ b)
Fig. 8. a) Design of a pad. b) If the boundary lines are not designed, the resulting pad is not square.
laterally reflected on the aluminium step that exposes the photoresist on a wider area. The use of a non reflective photoresist (1400 31 Dl) was necessary to avoid this effect. Because the
new resist has different properties, it was necessary to completely repeat the characterization
procedure on bare wafers. The aperture angle of the polariser was
= 20°, that corresponds
to a laser beam power of1.32 mW, as Po = 1.5 mW for the laser used. The line width is 9 pm.
The final processes are listed below:
Inversed antireflective photoresist (1400-31-Dl Shipley) spin coated: 30 s 6000 rpm.
First bake on hot plate: 105 °C 4 min.
Fig. 9. Short circuits caused by the relief parallel to the scanning (same experimental conditions
as those taken for Fig. 7, except a longer development time of 7 minutes).
hwrbean
area to expose
aTcaactuallyexpDsed
Fig. 10. Effect of topography on the exposed width.
Focused laser exposure using one scanning.
Second bake: 130 °C 6 min.
U.V. flood expose: I min.
Development: 4 min in pure developer (microposit developer Shipley).
Wet etching of aluminium in a Alu-etch bath.
IQIWNAL DEPHYSIQUED1 T.3,N°9, SE~ 1993 3l
Table I. Key parameters included in the model.
Parameter Value Influence
Wavelength fixed = 442 nm choice of the photoresist
of the Laser B earn
Power related to the angle of
°~~~~ ~°~~~ ~~~~° ~~ P°~~~~~
width on the resulting patterns (see fig.I)
P = Pocos28
Motion speed range : 0 to 10 mm s'l width
on the resulting patterns
of the Laser Bearn
Absorption coefficient choice of the photoresist
of ~° ~~ ~~~~~~~~~~
thickness
Threshold energy choice ofthe laser beam
of ~° ~~~~~~
Upon visual observation, the results were satisfactory (Fig. IIa). The lines are 8.9 ~Lm wide.
The pad boundaries are irregular due to the fact that it was designed as a square, not boarded by 4 rectangles (see Fig. 5, described in Section 2.2). For this particular application, it was
not a reliability issue. Then the connected transistor was electrically tested with a mechanical
probe station. The characteristics are given in Figure lib and are in a good agreement with the CAD predictions. No degradation of the prediffused elements and no significant added
resistance (less than 5$l) was provided by the interconnection processing.
3. Discussion
The aim of this approach is to establish a theoretical relationship between the set-up parameters of the system and the actual width of the final metal lines. These parameters, summarized in
Table I are the following ones: the wavelength of the laser beam which influences the choice of the photoresist, the power and the motion speed of the laser beam which influence the width of the resulting pattern, the light absorption coefficient of the photoresist and its threshold
energy which influence the choice of the photoresist's thickness as well as the set-up of the power and motion speed of the laser beam.
Figure 12 schematically shows the exposure of a thickness h of photoresist deposited on an
aluminium substrate. The laser beam comes perpendicularly along z axis on to the photoresist surface, described by z and y axes. The laser beam, considered as Gaussian, is focussed within the resist film, I.e. the beamwaist of the beam (point of smallest diameter on the laser beam)
is around the resist surface. This is due to the Focus Depth FD FD = 2 pm [Ii of the laser
beam.
The proposed model, founded on the Beer-Lambert law, uses a macroscopic approach.
a)
Ic(mA)
~_
~85[~~'~~
opA I
[
,
j
4
~(~([-
f 4
-/
-~ouA
-/
0000 2. 000
'icE 2000/div ( V)
b)
Fig. 11. Connected bipolar transistor. a) Optical view. b) Electrical characteristics.
Let us consider the Beer-Lambert's law:
where a is the absorbance of the photoresist. This absorbance changes during the exposure time [9]. an and ae are the value of the absorbance for the non-exposed photoresist (I.e. at
laserbean
x air
h photoresist
~ Alunfidwn
Fig. 12. Theoretical exposure of the resist on aluminium.
the beginning of the exposure) and for the completely exposed photoresist respectively; I(r, z)
is the light-intensity of the laser beam at a distance z of the beamwaist in a non-absorbing
medium and at a distance r (r~ = z~ + y~) from the axis of the laser beam. I(r, z) is given-by
the relation:
~ ~
I(r, z)
=
~~ ~~ ~ exp -2 ~~ (3)
xw w
where: Po is the power of the laser, is the aperture angle of the polariser. E [0,90°]. 2w is the diameter of the laser beam at the surface of the resist.
The power varies from 0 to Po according to the angle @. The above relation shows that the
spatial distribution of the light intensity in a plane perpendicular to the 0z axis decreases in
a Gaussian way. At a distance r
= 2w, the value of I(r, z) is considerably reduced (99.75% of
reduction).
RI " 0.064 and R2 " 0.998 are the reflection factors of the photoresist and of aluminium.
Because of the high value of R2, the three first reflections of the light on the substrate, within the photoresist, cannot be neglected. The light intensity incident upon the resist is:
1;n~(r,z)
= (I Ri)Ij(r,0) (4)
where I~(r, 0) is the light intensity dispensed on the photoresist surface.
Integrating along z axis the Lambert's law (Eq. 2), the quantity absorbed by the photoresist
is:
Iabs,0 " 1;nc eXp(-a/l) (5)
After the first reflection on the aluminium, the light beam is absorbed once again:
Iabs,1 " R2(1;nc ~abs,0 eXp(-a/l) (6)
Thking into account the two other reflections, the total intensity received by the photoresist is
given by:
~T " (1 RI eXp -all II + R2(1 ~XP(~°~l))
+RiR2(1 exp(-ah))~ + RIR((I exp(-ah))~] I~(r, 0) (~)
Interferences are neglegted: the standing waves effect within the layer gives an ondulation
on the edges ofthe etched line. As they affect only the edge shape, their influence on the mean width is neglecteable [10,11].
Table II. Physical parameters for the studied photoresists.
Photoresist 1400-27 1400-31-Di
Rj 0.064 0.064
h ~m I-I 1.5
cG ~m'~ 0.99 0,238
K~ calculated 0.740 0.683
c~~ pm-1 0.238 0.865
Kn calculated 0.709 0.364
E~ mJ.cm"~ 32 65
m 1.5 2
Let the laser beam move along the z axis at a speed called V. The absorbed energy during
time dt is: dEabs(z,y, z)
= IT(z,y, z) dt where dz
= Vdt. In order to simplify the model, and in accordance to the experimental results presented above, the following assumptions are
made: at any point to be exposed by the beam, the absorbance is a
= oe worth during the first third part of the laser beam passage, and then directly switches to a = an. Using equation (7)
and after integration along z, the expression above leads to the following formula:
The total energy absorbed must be the same (at least) as the threshold energy Ets necessary to sensitise the total thickness h of the photoresist. Thus, from equation (8) the width D of the line is then given by:
D
=
V5wIni ~~~ ~ (~~ ~ ~°~~ j
(10)
2 EtsU~
Kn, Ke and Ets depends on the materials used (substrate and photoresist). V was also set-up for most applications at I mm s~~ Po depends on the laser used. So, the adjustable parameters are w and @. If the laser beam is focused on the photoresist surface, w is the value of the beamwaist (99$l of applications). Finally, by varying @, and thus the power of the laser
beam, D can be choosen.
Two different photoresists were used. The usual one is the 1400-27 resist from Shipley.
However, because of topographical effects explained above, another kind of photoresist was
also tested (Shipley 1400-31-Dl). The parameters of these photoresists are listed on the Table II [12].
The variations of D as a function of the aperture angle are plotted in Figure 13, for the two types of photoresists. These variations are in very good accordance with the linewidth
obtained and measured, the differences are smaller than 10$~:
