Silicon-Micromachined Neurochips for In Vitro Studies of Cultured Neural Networks

Svetlana Tatic-Lucic, Yu-Chong Tai and John A. Wright Electrical Engineering, California Institute of Technology, Pasadena, CA 91125, U.S.A. Jerome Pine and Timothy Denison Physics, California Institute of Technology, Pasadena, CA 91125, U.S.A. ABSTRACT
The design, fabrication and testing of a novel neurochip structure for extracellular stimulation and recording of cultured neural networks are presented. The neurochip has 16 neuron wells, spaced 100 m apart, in a 4 x 4 array, which are fabricated on a 9 mm by 3 mm silicon membrane, 20-m-thick. The fabrication of the chips involves photolithography and micromachining on both sides of the membrane, which is located at the bottom of a 500-m-deep cavity. Experimentally, the biocompatibility has been demonstrated by culturing live neuron networks in the neurochips. Here, we describe the design, fabrication, and testing of a new chip structure, hereafter called the neurochip. The goal of the neurochip project is to achieve reliable and specific neuron-electrode connections for in vitro studies of cultured neural networks.

INTRODUCTION
The dynamics of long-term synaptic interactions is thought to be a key problem in understanding brain function. The study of synaptic plasticity is now possible in vitro, due to advances in cell culture, which allow specific neuron types to be removed from an animal and grown in a biochemical environment similar to their original habitat for periods of months. As a result, studies of synapse formation and neurite outgrowth are much easier to perform on cultured neurons. In general, it is not easy to achieve long-term, two-way electrical connection with the cells [1]. Moreover, difficulties may arise if one would like to perform experiments on cultured neural networks, where simultaneous electrical connection with many neurons is required. As for establishing electrical connection with cells, one way is to impale a neuron by a glass electrode filled with electrolyte. Another similar technique, so called patch-clamp technique, is using a thin glass pipette of the proper shape to tightly seal against a cell membrane [2]. Both techniques are extremely sensitive to voltage changes and highly specific in spatial resolution, but they can damage the cell during experiments. Also, it is extremely difficult to use more than two electrodes simultaneously, because of the bulky manipulators required to maneuver the electrodes into the correct position. Another category of methods is to use extracellular electrodes that capacitatively measure the cell potential. Silicon micromachining has been used to make such microdevices. For example, Najafi and Wise [3] have laid a valuable foundation for silicon multielectrode probes. Regehr, Pine and Rutledge [4] have demonstrated novel diving board electrodes for cultured neuron study. The major advantages of these silicon electrodes are their small sizes and biocompatibility. However, using these microdevices requires physical movement of the devices to, or near to, target neurons in order to form neuronmicrodevice connections. There are major drawbacks such as a lack of reliable and precise potential measurement due to a small signal-tonoise ratio and a poor spatial resolution.

Figure 1: Design and dimensions of a neurochip

Svetlana Tatic-Lucic, Yu-Chong Tai, John A. Wright, Jerome Pine and Timothy Denison, Silicon-Micromachined Neurochips for In Vitro Studies of Cultured Neural Networks, Technical Digest: International Conference on Solid-State Sensors and Actuators: Transducers 93, Yokohama, Japan, pp. 943-946, Nov. 1993.

DESIGN AND FABRICATION

The design of the neurochip is shown in Fig. 1. The neurochip has a 4 x 4 array of neuron wells (Fig. 2) situated at the center of a 9mm-long, 3-mm-wide and 20-m-thick silicon membrane. The minimum size of the membrane is determined by a GCA 4800 stepper in order to do autofocusing at the bottom of a 500-m-deep cavity. One side of the membrane has a 4 x 4 array of metal electrodes (Fig. 1a), and the other side has a corresponding grillwork array (Fig. 1b), which defines the neuron wells. Fig. 1c shows the schematic cross section of a neuron well. The depth of the neuron well, or the thickness of the membrane (20 m), is chosen to accommodate the size of the target rat superior cervical ganglion (SCG) neurons, about 15 m in diameter when fully grown. Fig. 3 shows an SEM photograph of a fabricated neuron well from the cavity side. Clearly seen is a heavily boron-doped silicon grillwork on top of the well and a circular gold electrode at the bottom.

The fabrication starts with epi-wafers with a 16-m lightly-doped layer on top of a 4-m heavily boron-doped layer (Fig. 4a). First, a composite layer of 180 nm LPCVD silicon nitride on top of 50 nm thermal oxide is formed. Photolithographic steps then pattern the nitride-oxide layer to define the openings (6 m in diameter) for the metal electrodes at the bottom of the neuron wells (Fig. 4b). Plasma and isotropic silicon etching steps then follow to etch through the nitride-oxide composite layer together with a 0.5 m recess into the silicon substrate. Next, a semirecessed LOCOS process [5] planarizes the surface and creates a 1 m oxide step around the electrode openings. The purpose of this is to ensure good electrical isolation of the electrode and to provide a tight seal to the cultured neuron in the well. The nitride is then stripped, and a 10 nm/200 nm/10 nm Cr/Au/Cr metalization is done using a lift-off process. This process forms the neuron-well electrodes as shown in Fig. 1a. This metalization is covered by a composite insulation layer of 0.5 m LTO and 1 m PECVD nitride (Fig. 4c). This combination is used for stress compensation. Next, opening of bonding pads and etching of alignment marks in the insulation layer are performed. The alignment marks are used later to do front-to-back alignment across a membrane that is micromachined in the subsequent steps. Windows on the back of the wafer are opened. The following EDP etching then forms the silicon membrane using the heavily boron-doped buried layer as an etch stop (Fig. 4d).

Figure 2: SEM of a 4 x 4 array of neuron wells at the bottom of a neurochip

Figure 3: SEM of a single neuron well 25 x 25 m2 at the top with 3 m circular gold electrode at the bottom

Figure 4: Cross section of the neurochip after some fabrication steps

Svetlana Tatic-Lucic, Yu-Chong Tai, John A. Wright, Jerome Pine and Timothy Denison, Silicon-Micromachined Neurochips for In Vitro Studies of Cultured Neural Networks, Technical Digest: International Conference on Solid-State Sensors and Actuators: Transducers 93, Yokohama, Japan, pp. 943-946, Nov. 1993.

During the same EDP etching step, silicon, including the borondoped buried layer, underneath the alignment mark is etched too. When the membrane is completed, square openings in the 20-m-thick silicon membrane are created, and, through the openings, alignment marks are visible for front-to-back alignment across the membrane. In fact, this technique allows alignment within 1 m, which is crucial for placing the electrode opening right at the center of each neuron well. Photolithographic steps on the cavity side of the membrane follow, and SF RIE etching is used to form grillwork in the heavily bo6 ron-doped layer. The neuron wells are formed by EDP etching all the way to the well electrodes on the front side of the wafer. The removal of pad oxide and chromium at the bottom of the wells then finishes the fabrication (Fig. 4e). Fig. 2 shows an SEM of a fabricated neuronwell array. Fig. 3 shows a close-up view of one well. Clearly seen in Fig. 3 is a 3 m circular gold electrode at the bottom defined by the semirecessed LOCOS steps.

EXPERIMENTAL RESULTS
The use of the neurochip is schematically shown in Fig. 5. First, an embryonic neuron is implanted into the well through the grillwork (Fig. 5a). Previously, the gold electrode was coated by platinum black, to decrease its impedance. The implanted neuron then grows in size, fills the well, forms a good seal with the bottom gold electrode, and grows processes out of the grillwork (Fig. 5b). The function of the grillwork then is to trap the grown neuron inside the well.
Figure 6: A live neural network formed by rat superior cervical ganglion neurons. Clearly seen are the processes growing out of the wells

Figure 5: a) Neuron just implanted in the well, b) the neuron increases in size and grows processes out of the well Experimentally, this use of our neurochips has been accomplished. A process has been developed to implant embryonic neurons into the wells without damaging them using micropipettes. This procedure of moving embryonic neurons in the well is shown in Fig. 6. Since the embryonic neurons are very delicate and fragile, moving them into the wells has been a major challenge in our experiments. Our process shows a survival rate of 75%, i.e., as much as three quarters of the neurons pushed into the wells can grow neurites out of the wells, which is accepted as the only evidence that they survive the implantation.

Figure 7: Major steps in the process of moving embryonic neurons into the wells a) Neuron sucked and held in pipette while being moved to a well b) Cell ejected from the pipette near a well c) Pusher positioned to move cell over the well. The pipette used for carrying the cell has been moved out of view d) Cell implanted in the well by a pusher

Svetlana Tatic-Lucic, Yu-Chong Tai, John A. Wright, Jerome Pine and Timothy Denison, Silicon-Micromachined Neurochips for In Vitro Studies of Cultured Neural Networks, Technical Digest: International Conference on Solid-State Sensors and Actuators: Transducers 93, Yokohama, Japan, pp. 943-946, Nov. 1993.

Moreover, the biocompatibility of the neurochip is evident that these implanted neurons further grow into live neural networks. As an example, Fig. 7 shows such a developed neural network cultured in our lab using rat SCG neurons. Currently, electrophysiological experiments such as stimulation and recording of the cultured neural networks are under way.

CONCLUSION
Neurochips for extracellular stimulation and recording of cultured neural networks have been fabricated. The neurochip has a 4 x 4 array of silicon wells, fabricated in a 20-m-thick silicon membrane. The double-sided alignment within 1 m of accuracy, as well as photolithography at the bottom of a 500-m-deep cavity have been developed for its fabrication. Development of live neuron networks in the neurochips demonstrates the biocompatibility of these neurochips. Acknowledgments - This work is supported by the National Institute of Health, Neural Prosthesis Program. We would like to thank Mr. Trevor Roper for his help during chip fabrication.

REFERENCES
[1] Wade G. Regehr (1988) Neuron-Microdevice Connections, Ph.D. dissertation, Division of Engineering, California Institute of Technology, pp. 2. [2] Erwin Neher and Bert Sakmann (1992) The Patch Clamp Technique, Scientific American, pp. 44-51. [3] Najafi, K. and K. D. Wise (1986) An Implantable Multielectrode Array with On-Chip Processing, IEEE J. Solid State Circ. 21, pp. 1035-1044. [4] W. G. Regehr, J. Pine and D. Rutledge (1988) A Chronic In Vitro Neuron-Microdevice Connection, IEEE Trans. on Biomed. Eng. 35, pp. 1023-1032. [5] N. Guillemot et al. (1987) A New Analytical Model of the Birds Beak, IEEE Transactions on Electron Devices, Vol. ED-34, No. 5, pp. 1033-1038.

Svetlana Tatic-Lucic, Yu-Chong Tai, John A. Wright, Jerome Pine and Timothy Denison, Silicon-Micromachined Neurochips for In Vitro Studies of Cultured Neural Networks, Technical Digest: International Conference on Solid-State Sensors and Actuators: Transducers 93, Yokohama, Japan, pp. 943-946, Nov. 1993.