ABSTRACT
Machine learning and signal processing on the edge are poised to influence our everyday lives with devices that will learn and infer from data generated by smart sensors and other devices for the Internet of Things. The next leap toward ubiquitous electronics requires increased energy efficiency of processors for specialized data-driven applications. Here, we show how an in-memory processor fabricated using a two-dimensional materials platform can potentially outperform its silicon counterparts in both standard and nontraditional Von Neumann architectures for artificial neural networks. We have fabricated a flash memory array with a two-dimensional channel using wafer-scale MoS2. Simulations and experiments show that the device can be scaled down to sub-micrometer channel length without any significant impact on its memory performance and that in simulation a reasonable memory window still exists at sub-50 nm channel lengths. Each device conductance in our circuit can be tuned with a 4-bit precision by closed-loop programming. Using our physical circuit, we demonstrate seven-segment digit display classification with a 91.5% accuracy with training performed ex situ and transferred from a host. Further simulations project that at a system level, the large memory arrays can perform AlexNet classification with an upper limit of 50 000 TOpS/W, potentially outperforming neural network integrated circuits based on double-poly CMOS technology.
ABSTRACT
The main requirement for using the Fluctuation Enhanced Sensing technique is the ability to perform low-frequency noise measurements. The portability of the measurement system is also a quite desirable feature not limited to this specific application. In this paper, an approach for the realization of a dual channel spectrum analyzer that is capable of exploring frequencies down to DC, although based on a USB sound card, is proposed. The lower frequency range of the input signals, which is outside the frequency range of the sound board, is upconverted to higher frequencies by means of a very simple modulation board. Then, the entire spectrum is reconstructed numerically by proper elaboration. With the exception of the modulation board, the approach we propose does not rely on any specific hardware. Thanks to the efficiency of the spectra estimation and reconstruction software, which is based on a public domain library, the system can be built on a low-cost computer single board computer, such as the Raspberry PI3. Moreover, when equipped with an optical TCP/IP link, it behaves as a compact spectrum analyzer that along with the device under test can be placed into a shielded environment, thus being isolated from external electromagnetic interferences.
ABSTRACT
In this paper an approach for designing a programmable, very low noise, high accuracy voltage source for biasing devices under test in low frequency noise measurements is proposed. The core of the system is a supercapacitor based two pole low pass filter used for filtering out the noise produced by a standard DA converter down to 100 mHz with an attenuation in excess of 40 dB. The high leakage current of the supercapacitors, however, introduces large DC errors that need to be compensated in order to obtain high accuracy as well as very low output noise. To this end, a proper circuit topology has been developed that allows to considerably reduce the effect of the supercapacitor leakage current on the DC response of the system while maintaining a very low level of output noise. With a proper design an output noise as low as the equivalent input voltage noise of the OP27 operational amplifier, used as the output buffer of the system, can be obtained with DC accuracies better that 0.05% up to the maximum output of 8 V. The expected performances of the proposed voltage source have been confirmed both by means of SPICE simulations and by means of measurements on actual prototypes. Turn on and stabilization times for the system are of the order of a few hundred seconds. These times are fully compatible with noise measurements down to 100 mHz, since measurement times of the order of several tens of minutes are required in any case in order to reduce the statistical error in the measured spectra down to an acceptable level.
ABSTRACT
Electrical noise sources are basic building blocks in many measurement and instrumentation applications and in communication systems. In this paper, we propose a quite simple topology for the realization of a programmable, wide bandwidth, white noise current source that requires only two resistors and one operational amplifier. We validate the proposed approach by means of SPICE simulations and demonstrate, by means of proper measurements, the capability of generating a flat current noise spectrum in a frequency range up to four decades from a few Hz up to 100 kHz.
ABSTRACT
The properties of a differential transconductance amplifier coupled with a four channel measurement system are exploited in order to reach a very high sensitivity in current noise measurements. In particular, it is demonstrated that, in proper conditions, the noise contributions coming from the active and passive devices that make up the transresistance amplifier can be virtually eliminated. Moreover, the proposed measurement method allows the evaluation of the impedance of the device under test from noise measurement data. Actual measurement results are also reported that demonstrate the effectiveness of the proposed approach.
