CHAPTER 3 LAYOUT DESCRIPTIONS AND EXPERIMENTAL RESULTS
3.3 EXPERIMENTAL RESULTS
Since most of the transistors in our chip operate in subthreshold region and the SPICE model for subthreshold circuit may not be accurate, the simulation results in the previous chapter might be different from the measurement results of the chip.
Therefore, it is necessary to compare the measurement results with SPICE simulation results. The measurement results are shown in the following.
First the measured transient response is shown. Fig. 29 shows the result of measurement when a periodic light source is incident on the chip. In the figure, overshooting and undershooting of PH1 could be observed clearly and so could that of the horizontal as simulated previously. To verify that the measured output of the chip is consistent with the HSPICE simulation results and the original CNN model, a normalized output of each part of the cell and simulation curves are plotted in the same drawing for comparison. The comparison is shown in Fig. 30. It could be found from the figure that the measured curves match the simulation curves qualitatively.
Fig. 31 shows the measurement of PH2 with different Vbias3. The measured time constant calculated from the rise and fall time of PH2 with different background illumination and Vbias3 is shown in Fig. 32. Since it is impossible to apply test signal directly to the input of PH2, the accurate time constant of PH2 is unavailable via chip measurement. The rise and fall time of the output of PH2 with supplying periodic signal are used to calculate the approximate time constant. The calculated results are shown in Fig. 32. As discuss in previous chapter, the time constant decreases with larger background illumination which could be evaluated roughly by IPH2. IPH2
represents the DC level of output of PH2. In the measurement, different background levels are produced by the aim of MN13 which adds an additional offset current to each cells input.
41
The spatial response is shown in Fig. 33. As mentioned in the previous
discussion, the smoothing or diffusion range could be tuned by controlling bias Vsm. A larger Vsm makes a wider smoothing range. The comparison of the measured spatial response and simulated response is shown in Fig. 34. The high contrast at the image edge is not as clear as that of simulation results, but the smoothing measurement result of horizontal closely match the simulation result.
Finally, the spatiotemporal patterns of the chip with subjected to periodic signal are shown in Fig. 35. The output of PH1 and horizontal are shown. The measured pattern is similar to HSPICE simulation results shown in the previous chapter. The output of PH1 spreads out laterally when the light is just incident. After some time, the output begins to contract in space gradually and reaches a steady-state value.
During the turn off transient, output current suddenly drops and returns to the steady state value. The pattern of horizontal appears similar .characteristic of spreading and contraction but the extent of spreading is wider. This spatiotemporal pattern is also similar to the biological measurement of real retinas shown in Fig. 36. The method of the biological measurement and recording are explained in [5].
42
Table II
Summary of characteristics of the proposed 32x32 retinal chip.
Process 0.35µm DPTM CMOS
Power Supply 3V
Resolution 32x32
Basic Cell Area 73.3µm x 73.3µm
Photo Sensor Area 40µm x 34µm
Number of Transistors in Basic Cell 39
Fill Factor 0.25
Total Area 2.6mm x 2.6 mm
Total Transistors of the Chip 41k
Power dissipation <30mW
43
VDDD CA1 VSSh VDDh CA2 CA2 CA2 CA2 VSSA
VSSD
VSSA BIAS1 BIAS2 BIAS3 BIAS4VDDh VSSh sm VP VB
VDDA
Fig. 24. Layout of the whole retinal chip. Main parts of the chip are labeled. Sensory array is in the center, the row and column address decoders are in the upper and left of the array respectively. Output buffer and address buffer are also labeled.
44
Photo-BJT
PH2 Horizontal
PH1
MP10
Fig. 25. Layout and floorplan of a basic cell of the sensory array. The four regions of main parts of the basic cell are surrounded by solid lines. The lower two sub region surrounded by dashed lines are phototransistor and PMOS capacitor MP10.
45
Column Decoder
Ro w De co de r
Output Buffer
Row Address Buffer
Column Address Buffer
Sensory Array
Fig. 26. Photograph of the whole retinal chip. Main parts of the chip are labeled.
Sensory array is in the center, the row and column address decoder are in the upper and left of the array respectively. Output buffer and address buffer are also labeled.
46
Photo-BJT PH2 Horizontal PH1
Fig. 27. Photograph of a basic cell. Only the photo-BJT region is not covered by metal.
47 Function Generator
Projector
100Hz
3V
Power Supply Oscilloscope
Chip
Lens LED
Fig. 28. The measurement setup diagram.
(a)
(b)
(c) (d)
Fig. 29. Measured transient response. Curve (a) the output of photo-input (b) PH2 (c) PH1 (d) horizontal. Under bias condition: Vbias1=1.52V, Vbias2=0.62V, Vbias3=1.11V, Vbias4=25mV, Vsm=1.26V.
48
(a)
(b)
(c)
Fig. 30. Comparison of the chip measured transient response to the model simulation results and HSPICE simulation results. (a) PH1, (b) PH2, (c) horizontal. The time units for model simulation and SPICE simulation are the same, 10µs/frame; time unit
for chip measurement is 20µs/frame.
49
Photo-input
Ph2
Ph1
Vbias3=1.2V
Vbias3=0.8V
Vbias3=0.5V
Fig. 31. Measured transient response with different Vbias3.. Under bias: Vbias1=1.5V, Vbias2=0.65V, Vbias4=20mV, Vsm=1.3V.
Fig. 32. Measured time constant calculated from rise and fall time of PH2. IPH2
represents the DC level of output of PH2.
50
(a)
(b)
(c)
Fig. 33. Measured spatial response. (a) Photo-input, (b) the output of horizontal, (c) PH1. Under bias condition: Vbias1=1.5V, Vbias2=0.65V, Vbias3=1.2V, Vbias4=312mV
51
(a)
(b)
(c)
Fig. 34. Comparison of the chip measured spatial response to the model simulation results and HSPICE simulation results. (a) Photo-input, (b) horizontal, (b) PH1.
52
(a)
(b)
Fig. 35. The measured spatiotemporal pattern of the chip. (a) PH1, (b) Horizontal.
The output current is normalized and then color-coded according to the bottom-right colorbar.
53
(a)
(b)
Fig. 36. The space-time pattern of (a) OFF bipolar cell, (b) horizontal cell layer measured from real retinas. Reprinted from [5]
54
CHAPTER 4
CONCLUSIONS AND FUTURE WORKS
4.1 CONCLUSIONS
In this thesis, a new retinal chip based on the biological model derived from the mammalian retina is proposed, analyzed, and measured. The proposed retinal chip is aim at reproducing the spatial and temporal response of real retina qualitatively. In spatial domain, a contrast enhancement is achieved while the overshooting and undershooting at the turn-on and turn-off transient could be observed in temporal domain. The main function of the chip is to reproducing these characteristics.
Through HSPICE simulation and chip measurement, the functions of the chip are verified. The overshooting and undershooting of the transient response and smoothing and contrast enhancement in spatial domain could be produced by the proposed chip.
The tunable parameter associated with the CNN model is the space constant of the horizontal. Varying bias voltage, the desired space constant could be obtained.
The chip is manufactured in 0.35µm double-poly triple-metal CMOS process.
The area of a basic retina cell is a square of 73.3µm, and the total area of the whole chip is approximately 2.6mm x 2.6 mm.
4.2 FUTURE WORKS
55
Though the function of the chip is verified to be correct, there is still room to improve. First, the time constant of real retina varies with different samples and that means the time constant of the model is not a exact but a tunable value. However in the proposed retinal chip, the time constant of each part could not be tuned. A new circuit offering a tunable time constant may be needed for potential applications.
Secondly, the cell area of the proposed retina is still too large with compared to that of a real retina cell. Shrinking the size of the cell may make some potential applications such as implantation of artificial retina in human body more practicable.
Finally, although the function of outer retina is realized by the proposed chip, the functions of other parts of retina model, inner retina, are not implemented. More designs are needed to realize the functions of the whole CNN model.
56
REFERENCES
[1] J. E. Dowling, The Retina: An Approachable Part of the Brain, Cambridge, MA:
Harvard University Press, 1987.
[2] F.S. Werblin, D. Roska, D. Bálya, Cs. Rekeczky, T. Roska, “Implementing a retinal visual language in CNN: a neuromorphic study,” Proc. ISCAS 2001, Vol. 2 pp.
333 –336, 2001.
[3] D. E. Haines, Fundamental Neuroscience 2nd ed. NY: Churchill Livingstone, 2002.
[4] B. Roska and F.S. Werblin, “Vertical Interactions across Ten Parallel Stacked Representations in Mammalian Retina,” Nature, Vol. 410, pp. 583-587, 2001.
[5] B. Roska, E. Nemeth, L. Orzo, F. Werblin, “Three Levels of Lateral Inhibition:
A Space-time Study of the Retina of the Tiger Salamander,” J. of Neuroscience:
1942-1951, 2000.
[6] Cs. Rekeczky, B. Roska, E. Nemeth, and F. S. Werblin, ”The network behind spatio-temporal patterns: building low-complexity retinal models in CNN based on morphology, pharmacology and physiology,” International Journal of Circuit Theory and Applications, Vol. 29, pp.197-239, 2001.
[7] D. Bálya, B. Roska, T. Roska, F. S. Werblin, “A CNN Framework for Modeling Parallel Processing in a Mammalian Retina”, International Journal of Circuit Theory and Applications, Vol. 30, pp. 363-393, 2002.
[8] T. Roska and L.O. Chua, “The CNN universal machine: an analogic array computer,”
IEEE Transactions on Circuits and Systems II: Analog and Digital Signal Processing, Vol. 40, March 1993.
57
Pages:163 - 173
[9] S. M. Sze, Semiconductor devices, physics and technology, New York :John Wiley & Sons, 1985
[10] S. Sedra, “The current conveyor: History and progress,” IEEE ISCAS Proc., pp.
1567-1571, 1989.
[11] C. Toumazou, F.J. Lidgey, and D.G. Haigh, Analogue IC design: the
current-mode approach, London, U.K.: Peregrinus on behalf of the Institution of Electrical Engineers, 1990.
[12] Alireza Moini, Vision chip, pp. 99-100, Boston: Kluwer Academic, 2000.
[13] P. R. Gray, P. J. Hurst, S. H. Lewis, R. G. Meyer, Analysis and design of analog integrated circuits, 4th ed., New York: John Wiley & Sons, 2001.
58
VITA
姓名: 楊文嘉
性別: 男
生日: 1979 年 1 月 31 日
出生地: 台灣新竹市
住址: 新竹市仁愛街 74 號 4 樓
學歷: 國立交通大學電子工程學系(1997-2002)
國立交通大學電子工程學系碩士班(2002-2004)
專長: 類比與混合式積體電路設計
Email: wjyang79@yahoo.com.tw
相關著作:
“The design of a bionic sensory chip based on the CNN model derived from the Mammalian retina”
Wen-Chia Yang; Li-Ju Lin; Chung-Yu Wu;