In-situ infrared dynamic sensing and computing array based on ferroelectric capacitors

Abstract

The present invention discloses an in-situ infrared dynamic sensing and computing integrated array based on ferroelectric capacitors, belonging to the field of semiconductor sensors. The array is composed of a plurality of pixels, each pixel is composed of a plurality of sensing and computing integrated devices, wherein the sensing and computing integrated devices at the same position in different pixels are connected in parallel by wiring, and the sensing and computing integrated devices are deep-grooved ferroelectric capacitors composed of a bottom electrode, a top electrode, and a ferroelectric dielectric material between the two electrodes on a deep-grooved insulating substrate. The ferroelectric dielectric material produces a pyroelectric charge response related to polarization intensity as the temperature changes. The present invention realizes zero-power in-situ infrared dynamic sensing, reduces system power consumption and delay overhead; and realizes weighted information perception, realizes four-quadrant multiplication of perception information and weights within a single device, and realizes complex in-sense computing functions.

Description

Technical Field

[0001] The present invention relates to the field of semiconductor sensors, and in particular to an in-situ infrared dynamic sensing and computing integrated array based on ferroelectric capacitors. Background Art

[0002] With the maturation of technologies like big data and AI, data volumes are experiencing explosive growth. Traditional cloud-based, computing-centric information processing architectures require the transmission of large amounts of data between edge devices and computing centers, resulting in high latency and significant power consumption. Furthermore, with the rapid development of the Internet of Things, the perception and recognition of dynamic objects with extremely low power consumption has become a core requirement for systems that wake up from random, sparse events, posing even more severe challenges to system power consumption.

[0003] Emerging in-sense computing architectures combine sensing units with information processing units, directly multiplying the sensor information by the unit's stored weights within the unit. This allows for information processing at the sensor end, alleviating the latency and power consumption issues associated with traditional cloud computing-based information transmission. One approach, based on connecting a sensing unit in series with a memory and computing unit (typically RRAM), utilizes Kirchhoff's principle to couple the sensor information with the unit's weights to achieve multiplication. However, this implementation incurs significant hardware overhead and, due to the presence of a DC path during sensing, consumes significant power. Other research is leveraging the sensing properties of two-dimensional materials, combined with the non-volatile gate control of charge storage transistors, to achieve in-sense computing functionality in a single device. However, this approach requires a high programming voltage, and the CMOS process compatibility and device reliability of two-dimensional materials are poor. Furthermore, existing in-sense computing solutions still require peripheral differential circuitry to process and recognize dynamic information, incurring additional hardware and power costs.

[0004] In recent years, researchers have discovered that ultrathin hafnium oxide (HfO2) exhibits ferroelectric properties under specific doping, stress, and annealing conditions. This material offers the advantages of CMOS process compatibility, scalability, and ultra-low power programmability, and is currently used in numerous storage and computing applications. Because the spontaneous polarization strength of ferroelectric materials varies with temperature, when the ambient temperature changes, the material in its stable state releases charge, a phenomenon known as pyroelectricity. This characteristic, which responds only to temperature changes and requires no additional power for sensing, holds great potential for applications in ultra-low-power dynamic sensing. Summary of the Invention

[0005] The purpose of the present invention is to propose an in-situ infrared dynamic sensing and computing integrated array based on ferroelectric capacitors, which can realize in-situ infrared dynamic sensing and computing functions with ultra-low power consumption.

[0006] The specific technical solutions of the present invention are as follows:

[0007] An in-situ infrared dynamic sensing and computing integrated array based on ferroelectric capacitors, characterized in that the array includes n pixels P1~P n The pixels are evenly distributed on the substrate in a matrix form, and each pixel is provided with m sensing and computing integrated devices arranged in a matrix form, wherein the i-th pixel has a sensing and computing integrated device SP i1 ~SP im The sensing and computing devices in the array are interconnected by back-end metal interconnects, where sensing and computing devices at the same position in different pixels are connected in parallel through the interconnects, and each group of parallel sensing and computing devices corresponds to an output port; the sensing and computing devices are deep-trench ferroelectric capacitors composed of a bottom electrode, a top electrode, and a ferroelectric dielectric material between the two electrodes on a deep-trench insulating substrate. The ferroelectric dielectric material generates a pyroelectric charge response as the temperature changes; in the programming state, the trained artificial neural network synaptic weights are written into the sensing and computing devices SP in the form of spontaneous polarization intensity through voltage pulses. 11 ~SP nm In, w im SP for the sensor-computing integrated device 11 ~SP nm The synaptic weights of the artificial neural network are used to sense the computing state. The entire array is in a short-circuit state. The changing infrared light intensity is converted into temperature changes and captured in the array. Different pixels perceive different temperature changes ΔT. i , and the sub-pixels storing different weights output weighted pyroelectric response charges ΔQ im =w im ΔT i The sensing and computing devices connected in parallel at the same position in different pixels output pyroelectric charges with weighted summation. Matrix-vector multiplication corresponding to the fully connected neural network realizes dynamic pattern perception function.

[0008] The top electrode is made of a light-transmitting conductive material, and the top electrode is preferably ITO;

[0009] The bottom electrode is made of metal and can provide sufficient stress during annealing to allow the ferroelectric dielectric layer to form ferroelectric crystals. The bottom electrode includes: TiN, TaN, Pt, Mo, Ru, etc.

[0010] The ferroelectric dielectric material is a conventional ferroelectric material such as perovskite ferroelectric (PZT, BFO, SBT), a ferroelectric polymer (P(VDF-TrFE)), or a novel ferroelectric material based on HfO2 that generates ferroelectricity under specific treatment (doping, stress, annealing, etc.).

[0011] The trench depth of the deep trench capacitor is preferably 100 nm, the trench diameter is preferably 50-150 nm according to the process, the thickness of the top and bottom electrodes is preferably 25-70 nm, and the thickness of the ferroelectric dielectric layer is preferably 8-15 nm.

[0012] The core of the present invention is to use ferroelectric capacitors to form a sensing and calculating integrated device with an in-situ infrared dynamic sensing function, wherein the ferroelectric capacitor uses the pyroelectricity of the ferroelectric dielectric material to realize in-situ dynamic infrared sensing without additional power consumption; further, the polarization intensity of the ferroelectric capacitor can be non-volatilely regulated by an external electric field, and dynamic infrared signal perception with non-volatile modulation can be realized. Wherein, the polarization intensity of the ferroelectric capacitor corresponds to the stored weight, and the different intensities and signs of the polarization intensity are combined to realize the multiplication of the dynamic sensing input and the stored weight; the present invention uses the above-mentioned ferroelectric capacitors to form an array based on pixels and sub-pixels through wiring, and each pixel is composed of a number of sensing and calculating integrated devices based on ferroelectric capacitors. In the entire array, the sensing and calculating integrated devices at the same position in different pixels are connected in parallel through wiring, and the output is unified, which can realize the multiplication and accumulation operation with weights in the sensor, thereby realizing the in-situ infrared dynamic sensing and calculating integrated function with ultra-low power consumption. The sensing and calculating integrated device provided by the present invention can realize in-situ dynamic sensing without the need for sensing power consumption, and can realize four-quadrant multiplication in which both the sensing input and the weight are signed.

[0013] Beneficial effects and corresponding principles of the present invention:

[0014] 1. Using the pyroelectric properties of ferroelectric materials for sensing can achieve in-situ dynamic sensing without the need for additional sensing power consumption.

[0015] Infrared detection is performed by utilizing the pyroelectric properties of ferroelectric dielectrics, that is, the change in their spontaneous polarization intensity when the temperature changes, thereby generating the characteristics of current and charge response. The response signal only appears when there is a change in temperature or infrared light intensity, and the sensing process is in a short-circuit or open-circuit state. No additional sensing power consumption is required, and zero-power in-situ dynamic sensing can be achieved.

[0016] 2. By utilizing the characteristic that the pyroelectric properties of ferroelectric dielectric materials change with the non-volatile polarization intensity, different device sensitivities can be achieved, thereby realizing a perception output with adjustable weights.

[0017] When a ferroelectric dielectric material is in a state of spontaneous polarization with different directions and intensities, the pyroelectric current or charge at the electrode will also vary with the direction and intensity of the spontaneous polarization under the same temperature change. Because the dielectric material is ferroelectric, the spontaneous polarization intensity can be non-volatilely controlled by an external electric field, thus achieving a sensory output with adjustable weights. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1Schematic diagram of an in-situ infrared dynamic sensing and computing integrated array based on ferroelectric capacitors according to an embodiment of the present invention;

[0019] Figure 2 2 is a schematic cross-sectional view of a sensing-computing integrated device prepared according to an embodiment of the present invention.

[0020] In the figure: 1—Substrate; 2—Pixels P1~P n ;3—Sensing and computing integrated device SP 11 ~SP nm

[0021] 4—top electrode (ITO); 5—ferroelectric dielectric layer (Hf 0.5 Zr 0.5 O2); 6—bottom electrode (TiN); 7—substrate (SiO2). DETAILED DESCRIPTION

[0022] The present invention will be further described below with reference to the accompanying drawings through embodiments.

[0023] like Figure 1 As shown, this embodiment provides an in-situ infrared dynamic sensing and computing integrated array based on ferroelectric capacitors, and the sensing and computing integrated device SP 11 ~SP nm The array combination of parallel vertical arrangement forms pixels P1~P n , the sensing and computing devices at the same position in different pixels are connected in parallel through the back-end metal interconnection lines. Figure 2 As shown, the in-situ infrared dynamic sensing and computing integrated device based on ferroelectric capacitors is composed of a top electrode, a bottom electrode and a ferroelectric dielectric layer on a deep groove substrate. The ferroelectric dielectric layer uses traditional ferroelectric materials such as perovskite ferroelectrics (PZT, BFO, SBT), ferroelectric polymers (P(VDF-TrFE)), or new ferroelectric materials based on HfO2 that produce ferroelectricity under specific treatment (doping, stress, annealing, etc.). The deep groove capacitor is preferably 100nm deep, and the deep groove diameter is preferably 50-150nm according to the process. The thickness of the electrode is preferably 25-70nm. The thickness of the ferroelectric dielectric layer is preferably 8-15nm.

[0024] The present invention also provides a manufacturing process for a sensing-computing integrated device based on a ferroelectric capacitor as follows:

[0025] (1) Defining the position and area of deep grooves on the SiO2 substrate by photolithography, and etching the grooves by wet etching or dry etching;

[0026] (2) preparing the bottom electrode in the deep trench by physical vapor deposition (PVD);

[0027] (3) Defining the bottom electrode contacts and connections by photolithography;

[0028] (4) growing a ferroelectric dielectric layer on the surface of the dielectric layer grown in step (2) by atomic layer deposition (ALD);

[0029] (5) continuing to grow the top electrode by physical vapor deposition (PVD);

[0030] (6) Define the capacitor area by photolithography and remove the remaining material by wet etching or dry etching.

[0031] material, exposing the bottom electrode;

[0032] (7) Through rapid thermal annealing (RTA) crystallization under certain conditions, the material produces ferroelectricity;

[0033] (8) Prepare contact electrodes.

[0034] In the programming state, the trained artificial neural network synaptic weights are written into the sensing and computing integrated device SP in the form of spontaneous polarization intensity through voltage pulses. 11 ~SP nm in; w im SP for the sensor-computing integrated device 11 ~SP nm The synaptic weights of the artificial neural network are used to sense the computing state. The entire array is in a short-circuit state. The changing infrared light intensity is converted into temperature changes and captured by the capacitors in the array. Different pixels perceive different temperature changes ΔT. i , and the sensing and computing integrated device that stores different weights outputs the weighted pyroelectric response charge ΔQ im =w im ΔT i The sensing and computing devices connected in parallel at the same position in different pixels output pyroelectric charges with weighted summation. Corresponding to the matrix-vector multiplication of the fully connected neural network, it realizes functions such as dynamic pattern perception.

[0035] The infrared light changes perceived in each pixel of the present invention are consistent, and each parallel sensing and computing device is pre-programmed to a specific polarization intensity, corresponding to different weights in response to changes in infrared light; the output of the array is different parallel sensing and computing device groups, corresponding to the weighted summation output of the infrared light dynamic signal. Multiple arrays are connected by metal wires to realize the function of a fully connected neural network.

[0036] The beneficial effects of the present invention are described with this embodiment:

[0037] 1. In the prior art, dynamic perception of environmental information requires additional modules such as differential circuits, which increases system power consumption and delay overhead. However, the present invention utilizes the pyroelectricity of ferroelectric materials to realize the function of the device responding only to signal changes. During the perception process, it is in a short-circuit state and does not require additional perception power consumption, thereby realizing device-level zero-power in-situ dynamic perception capability. The ferroelectric capacitor-based sensing and computing integrated device of the present invention utilizes pyroelectricity to realize zero-power in-situ infrared dynamic perception, reducing system power consumption and delay overhead.

[0038] 2. In the prior art, realizing the integrated sensing and computing function requires realizing modulated signal response, and signed perception requires redundant units to realize; however, the present invention utilizes the characteristic that the pyroelectric properties of ferroelectric materials change with their spontaneous polarization states to realize weighted perception of the device and non-volatile storage of weights. Since both the spontaneous polarization intensity and the perception signal have positive and negative signs, the four-quadrant multiplication computing capability is realized in a single device; the integrated sensing and computing device based on ferroelectric capacitors of the present invention utilizes the pyroelectricity regulated by the spontaneous polarization intensity to realize weighted information perception, realizes the four-quadrant multiplication of perception information and weights in a single device, and realizes complex in-sensing computing functions.

[0039] Finally, it should be noted that the purpose of disclosing the embodiments is to facilitate a further understanding of the present invention. However, those skilled in the art will appreciate that various substitutions and modifications are possible without departing from the spirit and scope of the present invention and the appended claims. Therefore, the present invention should not be limited to the contents disclosed in the embodiments; the scope of protection claimed by the present invention shall be determined by the scope defined in the claims.

Claims

1. An in-situ infrared dynamic sensing and computing integrated array based on ferroelectric capacitors, characterized in that: The array includes n pixels P1-P n The pixels are evenly distributed on the substrate in a matrix form, and each pixel is provided with m sensing and computing integrated devices arranged in a matrix form, wherein the i-th pixel has a sensing and computing integrated device SP i1 ~SP im The sensing and computing devices in the array are interconnected by back-end metal interconnects, where sensing and computing devices at the same position in different pixels are connected in parallel through the interconnects, and each group of parallel sensing and computing devices corresponds to an output port; the sensing and computing devices are deep-trench ferroelectric capacitors composed of a bottom electrode, a top electrode, and a ferroelectric dielectric material between the two electrodes on a deep-trench insulating substrate. The ferroelectric dielectric material generates a pyroelectric charge response as the temperature changes; in the programming state, the trained artificial neural network synaptic weights are written into the sensing and computing devices SP in the form of spontaneous polarization intensity through voltage pulses. 11 ~SP nm In, w im SP for the sensor-computing integrated device 11 ~SP nm The synaptic weights of the artificial neural network are used to sense the computing state. The entire array is in a short-circuit state. The changing infrared light intensity is converted into temperature changes and captured in the array. Different pixels perceive different temperature changes ΔT. i , and the sub-pixels storing different weights output weighted pyroelectric response charges ΔQ im =w im ΔT i The sensing and computing devices connected in parallel at the same position in different pixels output pyroelectric charges with weighted summation. Matrix-vector multiplication corresponding to the fully connected neural network realizes dynamic pattern perception function.

2. The in-situ infrared dynamic sensing and computing integrated array based on ferroelectric capacitors according to claim 1, characterized in that: The top electrode is made of a light-transmitting conductive material.

3. The in-situ infrared dynamic sensing and computing integrated array based on ferroelectric capacitors according to claim 1, characterized in that: The bottom electrode is made of TiN, TaN, Pt, Mo, or Ru.

4. The in-situ infrared dynamic sensing and computing integrated array based on ferroelectric capacitors according to claim 1, characterized in that: The ferroelectric dielectric material is a perovskite-type ferroelectric, a ferroelectric polymer, a traditional ferroelectric material, or a ferroelectric material based on HfO2 that generates ferroelectricity under doping, stress, and annealing conditions.

5. The in-situ infrared dynamic sensing and computing integrated array based on ferroelectric capacitors according to claim 1, characterized in that: The thickness of the top electrode or the bottom electrode ranges from 25 to 75 nm.

6. The in-situ infrared dynamic sensing and computing integrated array based on ferroelectric capacitors according to claim 1, characterized in that: The thickness of the ferroelectric dielectric material ranges from 8 to 15 nm.

Images