國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
Section 4.2 Electrical Properties Measurement Device Column 4.2.1 Device Fabrication
For the fabrication of electrodes, we need double layers of photoresist, S1813 and LOR5B. First, spin coating the LOR5B photoresist in 3000 rpm for 30 seconds, and prebake in 170C for 5min, then spin coating the S1813 photoresist in 3000 rpm for 30 seconds, and prebake in 120C for 3 minutes. Exposure to 7.7mw/cm2 UV light for 20 seconds, develop in MF - 319 for 40 seconds. Since the dissolve rate for LOR5B is much faster than S1813 in MF - 319, the photoresist layers will form undercut structure after development, which is needed for the metal deposition process later. 75s seems to be a good option for the etching process, and here's the result from surface profiler after etching, the depth of trenches is around 467 nm.
Fig 4.3 Surface condition after trench etched (467 nm in depth).
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
We use thermal evaporator for depositing Au as our electrode on SiO2 substrate. Since it's not easy to deposit gold on SiO2[18], we need to deposit a thin film of Ti for Au to adhere on it. In the end, we have seven pairs of electrodes with double layer Ti-Au structure, and the thickness of electrodes is 467-337=130 nm.
Fig 4.4 Surface condition after electrode deposited (337 nm in depth).
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
The photoresist used for making the channel is S1813, spin coat with 3000rpm for 30 seconds, and prebake in 120C for 3 minutes. Exposure in 7.7 mw/cm2 UV light for 11.5 s and develop in MF - 319 for 25 second.
Using ICP for the channel etching process. The depth of our channel should be around 1 µm, so we etch for 160 s – 1.06 µm.
Fig 4.5 Surface condition after channel etched (1.06 µm in depth).
Fig 4.6 Chip with channel and electrode.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
Fig 4.7 Chip with channel and electrode (10x).
Fig 4.8 Chip with channel and electrode (50x).
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
After the fabrication of electrode pairs and channel, now we need to seal the chip with PDMS. First, we cut PDMS into appropriate size and drill loading holes, then treat both PDMS and chip with oxygen plasma and put PDMS on the top of the chip, forming siloxane bond for sealing the channel.
Fig 4.9 Chip with channel and electrode after PDMS bonding.
‧
Column 4.2.2 Device Tests
After the fabrication, we have two main experiments to ensure if the device is functional. First, we check if the electric current can pass through the gap by conducting liquid buffer. Second, we check if we can obtain electric signal when target particles pass through those electrode gaps. Polystyrene beads are our first sample for the experiment since it can be monitored by fluorescence microscope, and are negatively charged, that we can transport it not only with mechanical force, such as pressure-driven flow, but also with electric field - which is a more controllable driving force than mechanical force. The buffer solution we use here is 1x TAE buffer - a common ionic buffer used in biological research - with 5% Tween 20 - making the buffer solution easier to flow through the microchannel.
Conduction Test
We load 1x TAE buffer with 5% Tween 20 - same with the buffer used for polystyrene beads - inside the channel and let the buffer only connect with two functional electrode pairs, that those electrode pairs without connection with TAE buffer can be seen as non-conductive comparison groups.
Fig 4.10 Conduction test electrode condition.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
Since for the measurement of small objects, the current need to be small enough – down to nanoamp or even lower - or it might destroy target during the measurement, a special circuit board is needed for the measurement. And now the circuit board is already fabricated by our co-worker, but it still needs to be calibrated. Now it is good enough for testing if the electrode pairs are conductive with the condition that the 1x TAE buffer is loaded into the channel.
Fig 4.11 Conduction test with 2 V bias.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
Fig 4.12 Conduction test with 2 V bias (reverse direction).
DC current level for electrode 1, 2 changed after the loading of TAE buffer, and DC current level for electrode 3, 5 still remain the same. Furthermore, different direction of voltage also results in the different direction of DC current change, which indicates that the electrode pairs are in conduction-state with our solution buffer.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
Sample Detection Tests
In these testing process, we start from smaller objects to larger objects. Smaller objects are harder for the electric detection, but it has a higher possibility of passing through the channel without blocking the channel, making the device not reusable. We used both electrical force and mechanical force for dragging sample through the channel.
(a) 200 nm fluorescence polystyrene beads
200 nm fluorescence polystyrene beads successfully flow through the channel, but as our expected, cannot receive any electric signal.
(a)
(b)
Fig 4.13 200 nm beads measurement.
(a) 200 nm bead pass No.1 electrode pair in t3 interval. (a) 200 nm bead pass No.1 electrode pair in t7 interval.
Fig 4.14 200 nm beads electric signal.
‧
500 nm fluorescence polystyrene beads successfully flow through the channel, but still cannot receive any electric signal. Something happened during the experiment, some of the testing beads stuck between electrode gaps when 1-volt measuring voltage applied.
t1 t2 t3 t4
Fig 4.15 Bead stuck between electrode gap.
After the measuring voltage changed to zero, some beads flow away from the electrode.
t5 t6 t7 t8
Fig 4.16 Bead released after measuring voltage turned off.
In the end of the experiment, we found out that there are more beads stuck between electrode gaps, and the DC electric signal level for both electrodes have slightly different with the level in the beginning of the experiment.
Fig 4.17 500 nm beads condition.
(a)In the beginning of experiment. (b)In the end of experiment.
Fig 4.18 DC current level comparison.
(a)In the beginning of experiment. (b)In the end of experiment.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
Section 4.3 Bacteria Image Processing System
We use the phase image of bacteria for the cell identification.
Fig 4.19 General phase image of bacteria.
Binarization is not enough for identification process, so several techniques need to be applied on this image before binarization.
Fig 4.20 Binarization only.
Original image, and binarzation applied with threshold value from high to low.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
The critical technique is the principal curvatures edge enhancement process, but the key idea for this technique is the sudden intensity change at the edge of bacteria, so any noises with high intensity inside the cell need to be removed or they might break the bacteria after the enhancement process.
Fig 4.21 Result without noise reduction.
Multiple smoothing techniques are needed for decreasing the impact from those noises.
Then obtain principal curvatures from noise reduction images.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
Fig 4.22 Image with noise and after noise reduction.
(a) Original phase image. (b) After noise reduction.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
Fig 4.23 Principal curvatures.
(a) (b) Eigenvalues of image’s Hessian matrix.
In Fig 4.22, (b) is better than (a) since the edge is much more clear and some cells are broken in (a), so we take image (b) for the next few steps.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
Fig 4.24 Binarization after edge enhancement.
It shows that the image quality after binarization is much better than before, all the bacterias are fully separated from the background, and there's no redundant connection between bacterias.
Since the size of E. coli is almost predictable, we can remove those dots which are much larger or smaller than we expect.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
Now we can do the identification using the command 'regionprops' in Matlab, and get image data of each bacterium one by one.
Fig 4.26 Applied regionprops function.
Fig 4.27 Identification result and comparison with original image.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
Now we apply this cell identification process to images in different conditions. Blue cells mean cells identified by this program.
Cell condition sets 1 and 2 are medium quality images, noises can be seen in the original image clearly.
(a) (b)
Fig 4.28 Identification result from cell condition sets 1.
(a) Cell condition sets 1.(b) Result.
(a) (b)
Fig 4.29 Identification result from cell condition sets 2.
(a) Cell condition sets 2. (b) Result.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
Cell condition sets 3 and 4 are low-quality images, noises can be seen in the original image clearly and there are some liquid stains in the background.
(a) (b)
Fig 4.30 Identification result from cell condition sets 3.
(a) Cell condition sets 3. (b) Result.
(a) (b)
Fig 4.31 Identification result from cell condition sets 4.
(a) Cell condition sets 4. (b) Result.
‧
Cell condition sets 5 is the most crowded image, also a medium quality image, noises can be seen in the original image clearly.
(a) (b)
Fig 4.32 Identification result from cell condition sets 5.
(a) Cell condition sets 5. (b) Result.
Table 4.2 Identification results.
In this table, 'number of cells' indicates the real quantity of cells in the image, 'counted cells' means the quantity of cells identified by cell segmentation program, 'error units' means numbers of units which are not our target but still being identified as a cell.
Sets\status (a) Number of Cells (b) Counted Cells (c) Error Units Identification Rate (b/a) Cell Counted per Error Unit Occurred (b/c)