Building an Analog Circuit Synapse for Deep Learning Neuromorphic Processing

1. Introduction

Neural networks are mathematical models that can be used to approximate functions. They work by adjusting the strengths of connections between neurons, called synaptic weights, based on the difference between the actual output and the desired output. This difference, called the error function, helps the network learn. Different learning rules are used in different contexts, such as control signals in control theory or policies in machine learning. Reinforcement learning (RL) methodologies are useful in tasks where scarcely available reward signals are provided, or well, the exact relationship between the system’s state vector $s_t$, the current action, $a_t$, and the reward signal is not clearly mapped into a function (i.e., a model-free system). Generative Adversarial Networks involve two neural networks that compete with each other for content generation. The goal of the first net is to generate new content (i.e., images, audio) indistinguishable from training data. The second network assesses the effectiveness of the first one by assigning a score to be maximized. DDPG, TD3, and Soft Actor-Critic neural architectures are advanced control algorithms that use two, three, and even four neural networks working together to produce the best results in control tasks. These algorithms are particularly useful when modeling the system and creating a proper policy is difficult. However, the training process can be computationally expensive, and conventional Von Neumann architectures are not optimal for this task because the storage and processing units are separated from each other, and additional circuitry is required to feed the processor with the necessary data. Spiking Neural Networks (SNN) attempt to replicate the cognitive mechanisms of biological brains by simulating the dynamics of neurons and synapses. This involves encoding and decoding information as spiking activity. Neuromorphic computing aims to create hardware that mimics this neuronal model, to achieve energy-efficient hardware with high throughput, embedded learning capabilities, and low energy consumption. The circuit implementation can be in the digital or analog domain. Digital neuromorphic computing involves developing digital hardware that can solve the differential equations of SNNs as quickly as possible. Examples of this type of hardware include Intel’s Loihi [1] and IBM’s Truenorth. This technology has already shown promising results regarding power efficiency and is a research platform compatible with current digital technologies. Digital to Analog Converters (DACs) and Analog to Digital Converters (ADCs) are used to quantify or binarize signals. However, using these converters always results in a quantization error, as larger binary words require larger DACs and ADCs. This implies that a greater number of quantization levels would lead to a smaller quantization error but larger circuit implementations without being reflected in better performance [2].

However, working entirely in the analog domain eliminates the quantization problem by treating information as circuit currents, voltages, charge, and resistance values. This approach allows for implementing neurons in analog counterparts, synapses with memristors, and additional circuitry in crossbar arrays. Using Kirchoff’s laws, values can be added instantaneously.

The conductance in each memristor enables in-memory computing and suppresses the von Neumann bottleneck. Using SNN models to assemble RL architectures can be counterproductive when executed on typical CPUs and GPUs. However, the same models can lead to high-performance and low-energy implementations if executed on neuromorphic devices, especially analog ones. However, as circuit analog design can be a challenging and iterative process, most frameworks/libraries or available tools for SNN are implemented in current digital technologies. For instance, Nest, SNN Torch, and Nengo [3,4,5] are software libraries that deploy SNN easily but are executed into current CPUs and GPUs. NengoFPGA is a Nengo extension to compile the network architecture into FPGA devices, which results in a digital neuromorphic hardware implementation. Therefore, most available tools and frameworks for SNN are currently implemented using existing technologies. Intel’s Lava is a compiler that uploads software-modeled SNN into Loihi chip. Both extensions, referred to as frameworks, result in digital neuromorphic implementations that are more efficient than running on von Neumann architectures. However, they are still digital. At [6], a population encoding framework for SNN is presented purely in the analog domain. The framework uses bandpass filters to distribute input signals into input currents for analog neurons evenly. However, storage and learning are not included in this framework. At [7], a Trainable Analog Block (TAB) is proposed that only considers encoding of signals. Information storage and obtention is left outside the scope of the study, as synapse values are computed offline and stored as binary words. To our knowledge, no end-to-end analog neuromorphic framework is available, including encoding, learning, and decoding in purely analog blocks.

This article presents a novel reward signal synapse circuit designed in the Skywater 130nm technological node to enable supervised learning into analog SNN circuits. The proposed structure enables a reward signal to switch between potentiation/depreciation of the synapse and spike reconformation to be implemented into a $n\times m$ fully interconnected neuron layers without having loss of power into the spikes, and also current decoupling, to supply the proper amount of current to the receptor neurons. Section 2 explains the modeling of SNN and the implementation of RSTDP learning rule dynamics in the synapse circuit. Section 3 describes the implementation of the memristor model in Verilog-A and the synapse circuit. Section 4 describes the neuron CMOS model used to test the synapse. A $2\times 1$ neuron network structure is tested in simulation, demonstrating adequate learning capabilities. Section 5 consists of discussion, conclusions, and future work.

2. Preeliminars

Now, lets proceed by describing briefly the system dynamics of neurons, synapses and learning algorithms for SNN´s, in order to understand the resulting circuitry, down further the text.

2.1. Spiking Neural Networks

The behavior of biological brains, including the interactions between synapses and neurons, can be mathematically modeled and recreated using equivalent circuitry. One such example is the biological neuron, which has various models that range from biologically plausible but computationally expensive (such as the Hodgkin and Huxley model [8]) to simplified yet reasonably accurate models. The Leaky Integrate and Fire (LIF) neuron model simplifies neuron dynamics by approximating the neuron’s membrane as a switched low-pass filter:

$${\tau }_m\frac{d{v}_m\left(t\right)}{dt}=E_L-v_m\left(t\right)+R_m{I}_{ext}\left(t\right)$$

(1)

$$v_{out}=\left\{\begin{array}{c}{v}_{spk}·\delta \left(t\right),\forall v_m=v_{th}\hfill \\ 0,\forall v_m<v_{th}\hfill \end{array}\right.$$

(2)

In Equation (1), $v_m\left(t\right)$ represents the membrane’s voltage, which has certain membrane’s resistance $R_m$ and capacitance $C_m$. The temporal charging constant of the neuron ${\tau }_m=R_m{C}_m$ imposes a charging/discharging rate as a function of an input excitation current $I_{ext}$, starting from a resting potential $E_L$. When $v_m$ overpasses certain threshold voltage $v_{th}$, the neuron emits a spike of amplitude $v_{spk}$, being $\delta \left(t\right)$ the Dirac delta function. As described at [9], by solving the differential equation in the time interval it takes to the neuron to $v_m=v_{th}$ and considering the frequency definition, a function which relates $I_{ext}$ with the output spiking frequency can be obtained as:

$$f_{spk}\left(I_{ext}\right)=\frac{1}{{\tau }_{m}ln\left(\frac{v_{th}-E_L-R_m{I}_{ext}}{R_m{I}_{ext}}\right)}$$

(3)

The resulting graph is called tuning curve [7] and depicts the sensibility of the neurons against an excitatory signal. By varying $C_M,R_m$, different tuning curves, i.e., spike responses, can be obtained for neurons in the same layer. (See Figure 1). For instance, a larger value for $C_m$ will make the neuron take more time to charge, reducing the spike output frequency and leading to a different tuning curve. This feature can encode input signals into spiking activity by letting neurons in the same layer have different spike responses for the same input signal (i.e., Population Encoding).

2.2. R-STDP Learning Rule

Spike-Timing-Dependent Plasticity (STDP) describes Hebbian learning as neurons that fire together, wire together. Given a couple of neurons interconnected through a synapse, the ability to conduct the spikes is controlled by a synaptic weight $w\in [w_{min},w_{max}]$. The neuron that emits a spike at time $t_i$ is denoted as the pre-synaptic neuron $N_i$, making the synapse increase its conductance value by a certain differential amount $\mathsf{\Delta }w$. A spike of current resulting from the convolution of the spike’s voltage through the synapse is produced and fed a receptor post-synaptic neuron $N_j$. Each spike contributes to the membrane’s voltage of $N_j$ until it emits a spike at time $t_j$, becoming then the pre-synaptic neuron, and setting ${\mathsf{\Delta }}_t=t_i-t_j$ as the time difference between spikes. $\mathsf{\Delta }w$ is then defined as:

$$\mathsf{\Delta }w\left(\mathsf{\Delta }t\right)=\left\{\begin{array}{c}{A}_{+}{e}^{\frac{-{\mathsf{\Delta }}_t}{{\tau }_{+}}},\forall {\mathsf{\Delta }}_t\ge 0\hfill \\ -A_{-}{e}^{\frac{{\mathsf{\Delta }}_t}{{\tau }_{-}}},\forall {\mathsf{\Delta }}_t<0\hfill \end{array}\right.$$

(4)

For each spike, the synaptic weight will be modified by a learning rate of $A_{+},A_{-}$, multiplied by an exponential decay defined by ${\tau }_{+},{\tau }_{-}$, respectively. As ${\mathsf{\Delta }}_t\to 0$, the change in the synaptic weight is bigger. Figure 2b) shows the characteristic graph of STDP, showing that for presynaptic spikes (${\mathsf{\Delta }}_t\le 0$), the synapse gets Long Term Potentiation (LTP), while for post-synaptic spikes (i.e., ${\mathsf{\Delta }}_t\ge 0$), the synapse suffers with Long Term Depreciation (LTD). The resulting plasticity rule models how the synaptic weight is modified, considering only the spiking activity. According to [10,11,12], a global reward signal R is introduced to model neuromodulatory signals. Setting $R\in [-1,1]$, Equation (4) is changed then to:

$$\mathsf{\Delta }w\left(\mathsf{\Delta }t\right)=\left\{\begin{array}{c}R\times A_{+}{e}^{\frac{-{\mathsf{\Delta }}_t}{{\tau }_{+}}},\forall {\mathsf{\Delta }}_t\ge 0\hfill \\ -R\times A_{-}{e}^{\frac{{\mathsf{\Delta }}_t}{{\tau }_{-}}},\forall {\mathsf{\Delta }}_t<0\hfill \end{array}\right.$$

(5)

Figure 2c) shows the role of the reward signal $R=-1$, inverting the role of presynaptic and postsynaptic spikes. Presynaptic spikes now lead to LTD, while postsynaptic spikes lead to LTP. This is the opposite of STDP (i.e., $R=1$). Notice when $R=0$, learning (modification of the synaptic weights) gets deactivated, as $\mathsf{\Delta }w=0$.

3. Materials and Methods

This section describes the necessary circuitry assembled to emulate the models for synapses, neurons, and learning rules.

3.1. Memristor Device

A Resistive Random Access Memory (RRAM) device consists of a top and bottom metal electrodes (TE and BE, respectively), enclosing a metal-oxide switching layer, forming a metal insulator metal (MIM) structure. A conductive filament starts to be formed with oxygen vacancies when current flows through the device. The distance from the tip of the filament to the opposite bottom electrode is called gap g. Notice that the length of the filament s is complementary, as the thickness of the oxide layer $t_{ox}=g+s$, as it can be seen in Figure 3a). Reverse current increases g while the device’s resistance increases, and vice-versa. Skywater’s 130nm fabrication process (see [13]), incorporates memristor cells produced between the Metal 1 and Metal 2 layers and can be made using materials that exhibit memristive behavior, such as titanium dioxide (TiO2), hafnium dioxide (HfO2), or other comparable materials based on transition metal oxides.

The RRAM can store bits of information by switching the memristor resistance value $R_{mem}$ between a Low Resistance State (LRS) and a High Resistance State (HRS). However, this work’s intention is to use the whole range of resistance available to store the synaptic weights by directly representing $w_{min}$ with $HRS$, $w_{max}$ with $LRS$ and any continuous value in-between. UC Berkeley’s model [14] defines the internal state of the memristor as an extra node in the tip of the formed filament. The memristor dynamics is described by the current between TE and BE electrode $i_{TE,BE}$, the rate of growth of the gap $\dot{g}$, and the local field enhancement factor $\gamma$:

$$i_{TE,BE}\left(t\right)=I_0·exp(-g/g_0)·sinh(v_{TE,BE}/V_0)$$

(6)

$$\frac{d}{dt}g\left(t\right)=-{\nu }_0·exp(-\frac{E_a}{V_T})·sinh\left(\frac{v_{TE,BE}·\gamma ·a_0}{t_{ox}{V}_T}\right)$$

(7)

$$\gamma ={\gamma }_0-\beta ·g^3$$

(8)

where $v_{TE,BE}$ is the voltage between TE and BE, $t_{ox}$ is the thickness of the oxide separating TE and BE, $a_0$ is the atomic distance, and $I_0,V_0,g_0,V_T,{\nu }_0,{\gamma }_0,\beta$ are fitting parameters obtained from measurements of the manufactured memristor device [15].

Then, to introduce a device model as a circuit component in a simulation environment, the user can a) use SPICE code to reflect the memristor model, or b) describe the device dynamics using Verilog-A, and then use the simulator’s compiler, being the latter option the standard. Simulation results described at [16,17,18], shows succesfull transitionary simulations, using a 1T1R (one-transitor-one-resistor) and 1C1R (one-capacitor-one-resistor) configurations, using the Verilog-A code provided at [19], compiled and simulated with Xyce/ADMS [20] software. However, over these simulations using pulse excitation signals, while they report successful decay in the memristance value, they do not report how the memristance value goes up again by applying pulses in the opposite direction. We could not reproduce the mentioned behavior to the best of our efforts.

It can be noticed on the provided code that the Euler integration method is described, alongside the model, by request to the simulator engine the absolute simulation time at each timestep with Verilog-A directives such as initial-timestep or absolute-time. While this works for .tran simulations, this model description will fail for .op, .dc simulations, where time is not involved or will lead to convergence simulation issues. These and other bad practices are described in detail by the UC Berkeley’s team article [14]. They provide insight into how to model devices with hysteresis in Verilog-A by properly:

• Defining a couple of differential equations, $f_1(v_{TE,BE},g)$ (Equation (6)) to describe the current from between TE and BE terminals, and a second function $f_2(v_{TE,BE},g))$ (Equation (7)) to describe the rate of growth/shrinkage of the width of the filament.
• Defining TE, BE and the tip of the filament (i.e., g) as electric nodes in Verilog-A. As each node in an electrical circuit possesses properties (Voltage, Currents, Magnetic Flow, and charge), the compiler knows how to compute the current from the tip of the filament to $BE$, by using $f_2(v_{TE,BE},g))$
• Providing alternative functions implementations for exp(), sinh(), to limit the maximum slope these can reach between the past and the next timestep. Several simulator engines use dynamic timestep selection for faster simulation periods and convergence issues. Of course, this limits the minimum timestep a simulator can use but avoids convergence issues or extended execution periods.
• Avoiding the usage of if-then-else statements to set the boundaries for the thickness of the filament. Instead, use a smooth, differentiable version of the unit step function.

This article uses the memristor Verilog-A implementation methodology described at [14] but replaces the manufacturing parameters found at [16,17]. Figure 3a) shows a memristor testbench where a triangular signal from $-2V$ to $2V$ is applied, resulting in the Lissajous curve (I-V graph) (Figure 4a), with the typical hysteresis characteristics from memristors, reflecting the $V_{on}\approx 0.9V$ and $V_{off}\approx -0.6$ threshold voltages to increase/decrease the resistance in the device. Figure 4b) shows the thickness of the filament, which lies between$4.9nm$ to $3.3nm$. On a second testbench depicted in Figure (Figure 3b), a 1T1M (one transistor-one memristor) setup where squared pulses ((Figure 4a) are applied first at $BE$, then at $TE$ to foster the resistance value exploration, reflecting a proper evolution from LRS to HRS and backward, showing the appropriate previously reported memristance values of the device, this is, $R_{mem}\in [10k\Omega ,3.3M\Omega ]$ (Figure 4b). The current flown through the memristor goes from $-200\mu A$ to $100\mu A$, matching the obtained Lissajous curve in the previous testbench (Figure 4c). The resulting code is available at our Github repository [21], compiled by the OpenVAF tool [22], and simulation results were obtained using the Ngspice simulation engine [23].

3.2. R-STDP Circuit Implementation

Figure 6 shows a 4T1M (four transitors-one memristor) cell replacing the 1T1M cell to manage the current flow of the memristor. Proposed at [24], the structure pretends to invert the current flow according to a reward voltage signal $V_R\in [0,1.8V]$. When $V_R=1.8V$, transistors $M1,M2$ are enabled, and $M3,M4$ disabled. If $V_{pre}>V_{post}$, the current then will flow from BE of the memristor towards TE. However, when $V_R=0V$, $M2,M3$ are disabled, and $M1,M4$ are enabled, yielding the current direction from TE to BE. The current direction determines whether the memristor´s resistance increases or decreases. On Figure 7a) testbench signals with triangular shapes are applied with a delay of $5\mu s$. Two scenarios are presented: In the first scenario, $V_R=0$ for the entire simulation, while in the second scenario, $V_R$ flips from $0V$ to $1.8V$ at $100\mu s$. Notice at Figure 7c), Figure 7d) the voltage difference ($V_{mem}=V_{BE}-V_{TE}$) is shown for both scenarios, overpassing the memristor threshold voltage for potentiation. However, when the reward signal is flipped, $V_{mem}$ overpasses the memristor value but in the opposite direction. Notice also at Figure 7e), Figure 7f) the magnitude of the spike currents the memristor delivers, given by the $M1-M4$ geometry $W/L$, which in this case, where selected to provide symmetrical pulses of current. However, these aspect ratios can be selected to foster asymmetric STDP curves.

3.3. Adding Spike Reconformation and Current Decoupling to the Synapse

Now consider two fully-interconnected neuron layers, with N and M neurons needing $N\times M$ synaptic connections. When the $i-th$ neuron of the first layer emits a spike, it should be able to provide enough power for the M post-synaptic neurons. Moreover, when the $j-th$ neuron in the second layer emits a spike, it must provide enough power to the N post-synaptic neurons. Consider then the schematic in Figure 8. Notice that the $4T1R$ R-STDP structure of the previous section is embedded inside this new $11T1R$ structure, supporting the polarity switch according to the arrival of spikes. The port labeled as $V_{out,pre}$ activates transistor $M6,M7$. making $V_{pre}>V_{post}$. On the other side, the port labeled as $V_{out,post}$ enables transistors $M5,M8$, setting $V_{post}>V_{pre}$. Then, four scenarios, depicted at Figure 9) emerge:

• $V_{pre}>V_{post}$$R=1.8V$. When a presynaptic spike arrives and the reward signal is on. This routes the current from BE to TE in the memristor, yielding to LTD;
• $V_{pre}>V_{post}$$R=-1.8V$. Due to the reward signal being negative, the same spike train that should produce LTD now produces LTP, as the current flows from TE to BE;
• $V_{post}>V_{pre}$$R=1.8V$. Postsynaptic spikes with reward signal on, the current flows from TE to BE, producing LTP;
• $V_{post}>V_{pre}$$R=-1.8V$. Postsynaptic spikes with a reward signal off, the current flows from BE to TE, producing LTD;

As the input spikes are pointing toward the transistor’s gates, no current is provided by the neurons. Instead, each synapse only receives the trigger signal (a spike, with amplitude larger than the threshold value of the transistors), and provides enough current straight from the power source, instead of the output node $v_{out}$ of each neuron. Regarding the upper part of the circuit, it’s important to note whether the spike was presynaptic or postsynaptic; the current that flows through transistor $I_{M10}$ always travels from source to drain. Additionally, this current is the same that flows through the memristor, regardless of its polarity. Transistors $M9,M11$ then serve as current mirrors of $I_{M10}$. When the postsynaptic neuron fires, current is delivered to the presynaptic neuron by $I_{pre}$. Then, when a presynaptic spike arrives, $I_{post}$ feeds the post-synaptic neuron. $I_{post}$ and $I_{pre}$ are defined as:

$$I_{post}=\frac{{(W/L)}_{11}}{{(W/L)}_{10}}{I}_{M10}$$

(9)

$$I_{pre}=\frac{{(W/L)}_9}{{(W/L)}_{10}}{I}_{M10}$$

(10)

As mentioned in the previous sections, the resistance range of the memristor allows it to provide at most $200uA$. However, the input current range of the neuron may differ from the input current ranges the memristor can provide for the same $vdd$. Therefore, Equations (9) and (10) enable regulation of the current contribution for each spike.

3.4. Neuron circuit

Figure 10 depicts the neuron model used in this work, based on the original design by [25], but with some modifications for the avoidance of the output spike to be fed into the same neuron, as seen as [24]. Transistors $M3,M4,M5$ emulate a thyristor with hysteresis by harnessing the fact that PMOS and NMOS have different threshold voltage values (i.e., $v_{THN}\ne v_{THP}$). The circuit dynamics can be described as follows:

• An external input current excitation $I_{ext}$ arrives through $M10$ (PMOS), enabled at start. $M1$ is set as a diode.
• $C_1$ charges for each incoming spike, increasing the voltage at node $V_m$.
• A leaky current $I_{leak}$ is flowing thourgh $M2$ at all times. If no further incoming electrical impulses are received, the neuron will lose all of its electrical charge. $V_{b1}$ defines $I_{leak}$.
• When $V_m\approx 1.5V$, $V_g\approx 0.75V$, which is the threshold voltage for the $M5$ NMOS device, enabling the $C_1$ charge to flow through $M4$ and $M5$.
• $M7$ also turns on, enabling current to flow and making voltage at $\overline{V_{out}}$ drops. At the same time, $V_g>0$, turning off transistor $M10$, disabling current integration for the neuron.
• As ${\overline{v}}_{out}$ drops, $v_{out}$ rises, as $M8-M9$ works as an inverter. $V_{b2}$ controls the current of the transistor $M6$, and conforming the width of the spike. The node $v_{out}$ provides the final output spike, which can be fed to subsequent synapses.
• $M10$ acts as a controlled diode, blocking any current from $I_{ext}$ when the neuron is spiking.

Figure 11 shows the neuron’s spiking activity for an input step signal, which rises each $50\mu s$ a step of $50nA$. It can be observed that the frequency increases as more current is added. The amplitude of the output spikes is set as $v_{out}=1.8V$, but it can be set differently, according to the synapse needs, while the thickness of the spike is approximately $1ms$. The maximum reached spike frequency is $f_{out}\approx 140kHz$, for $I_{ext}\approx 160nA$. The neuron output voltage $v_{out}$ remains on $1.8V$ for bigger input currents. The final design then needs ten transistors. However, notice $M8-M9$ can be removed by considering $V_g$ as the output voltage node. The geometry of $M5$ can be reshaped to regulate its current, defining then the width of the output spike.

4. Results

The testbench shown in Figure 12 intends to test all the capabilities of the proposed synapse by using a 2-1 neuron network array, using then two $11T1R$ synapses to interconnect the first layer with the second. During the first half of the simulation, the neurons N1 and N2 in the input layer receive testing excitation currents for each neuron, supported by current mirrors. The neurons at the first layer spike at different rates, as $I_{ext,N1}>I_{ext,N2}$. In the second half of the simulation, the neuron N3 in the output layer receives an excitation current $I_{ext,N3}$, while the current mirrors for N1 and N2 get deactivated. This should result in Long-Term Potentiation (LTP) for the first half and Long-Term Depression (LTD) in the second half of the execution. However, the reward signal R is set to switch from 1.8V to -1.8V at each quarter of the simulation, leading to the four cases previously described.

Figure 13d shows the evolution of the simulation results. In the first quarter, the first case occurs, showing how the thickness of the filament s in the memristor decreases at distinct rates. Remember, as the thickness decreases, so does the conductivity (i.e., LTP is produced). In the second quarter, the same spiking activity led to the opposite effect in the filament, as the reward signal R went from $1.8V$ to $-1.8V$, leading the thickness to $4.9nm$ (i.e., LTD is then produced). In the third quarter, $N3$ spikes, and $N1,N2$ ceases to receive excitation current from the current mirrors. With $R=1.8V$, the filament should increase in size; however, as its value is already at the maximum, it stays at $4.9nm$. Finally, R flips again to $R=-1.8V$, and the spikes of $N3$ decrease the filament, obtaining the opposite behavior. Notice the neural activity (spikes), byproduct of current integration of incoming spikes of other neurons (See Figure 13c).

The whole circuit is implemented using Xschem and Ngspice. The geometries for each transistor used for this testbench are described in Table 1. The final list of archives is available at our Github repository [21].

5. Discussion

Next, we point towards some considerations into the obtained circuitry, how this device can be manufactured, and new opportunities of research which emerge from this work.

5.1. Regarding the synapse circutry

The simulation took around $10s-15s$ for a timestep of $\mathsf{\Delta }t=100ns$ to the implementations, without sudden crashes or singular matrix values in simulation runtime. This is thanks to the well-posed memristor Verilog-A code. This will enable the simulation of larger Deep Neural SNNs purely implemented on an integrated circuit. Also, at [24], while they do implement a GAN network, some blocks like the memristor Generic VTEAM memristor mathematical model [26] are not manufacturable. All the presented architecture has feasible manufacturing using the Skywater 130nm process node.

5.2. Future Work towards Tailor-Made Neuromorphic Computing

The presented blocks then work properly once assembled, leading to the research into how to assemble larger structures. Drawing a schematic of a bigger network with hundreds/thousands of neurons per layer may be problematic, even using the neurons and synapses as sub-blocks. Not to mention the layout. Figure 14a) shows the resulting layout of the $11T1R$ synapse structure, comparable in size with the LIF neuron at Figure 14b), drawn in the Magic VLSI EDA software [27]. Notice that both structures have internal geometries that might be parameterizable. For instance, the length of the current mirrors for $I_{pre},I_{post}$ in the synapse, or well, the capacitance value $C1$ for the neuron. As future work, the authors consider that our efforts should focus on automatizing the SPICE files (using programming libraries like PySPICE) for simulation and the GDS files (like implementing PCELLs in TCL scripting) for manufacturing. This will enable research into implement deep neural networks architecture purely in analog hardware.

6. Conclusions

In this article, a proper implementation in Verilog-A of the Skywater130 reram model is presented, enabling the simulation of the behavior reported in the characterization of the physical device while allowing simulations in shorter periods without convergence issues. Then, a $11T1R$ synapse structure, which uses the memristor inside, is presented. This structure not only enables reward modulation for STDP but also decouples the current feeding from the neuron and successfully transfers the power duties to the synapse. The output spikes of the neuron then do not provide current for the memristor but only provide the signaling to enable the flow into one direction or the other.