Home on Fabrizio Ottati

Neural inference at the frontier of energy, space and time - NorthPole, IBM

Tue, 26 Dec 2023 00:00:00 +0000

NorthPole is the new shiny artificial intelligence (AI) accelerator developed by IBM.

NorthPole, an architecture and a programming model for neural inference, reimagines (Fig. 1) the interaction between compute and memory by embodying 10 interrelated, synergistic axioms that build on brain-inspired computing.

The creators claim that NorthPole design has been driven by 10 axioms inspired by human biology. We will use them as guidelines to analyze the paper, keeping the same outline as the original article.

Neural inference at the frontier of energy, space and time - NorthPole, IBM

Sat, 25 Nov 2023 00:00:00 +0000

NorthPole is the new shiny artificial intelligence (AI) accelerator developed by IBM. The creators claim that NorthPole design has been driven by 10 axioms inspired by human biology. We will use them as guidelines to analyze the paper.

The outline of this blog post is the same as the original article.

Axiomatic design

NorthPole, an architecture and a programming model for neural inference, reimagines (Fig. 1) the interaction between compute and memory by embodying 10 interrelated, synergistic axioms that build on brain-inspired computing.

Installing Vitis HLS and Vivado on Fedora 37

Thu, 25 May 2023 00:00:00 +0000

Fedora 37 is not officially supported by Xilinx (or AMD, I should say). Hence, I had to use some workarounds to install it on my work machine.

These instructions have been tested for the 2022.2 and 2023.1 versions.

Why Vitis/Vivado?

Vitis HLS and Vivado are freeware! This means that the tool itself is not open source, but anyone can download it and use it! I use it in my research and I find it useful to be able to run some small testbenches on my local machine without ssh-ing to the lab server.

Bits of Chips | TrueNorth

Sun, 26 Mar 2023 00:00:00 +0000

Why do we want to emulate the brain?

If you have ever read an article on neuromorphic computing, you might have noticed that in the introduction of each of these there is the same statement: “The brain is much powerful than any AI machine when it comes to cognitive tasks but it runs on a 10W power budget!”. This is absolutely true: neurons in the brain communicate among each other by means of spikes, which are short voltage pulses that propagate from one neuron to the other. The average spiking activity is estimated to be around 10Hz (i.e. a spike every 100ms). This yields very low processing power consumption, since the activity in the brain results to be really sparse (at least, this is the hypothesis).

Efficient compression for event-based data

Tue, 28 Feb 2023 00:00:00 +0000

Efficient compression for event-based data

Datasets grow larger in size

As neuromorphic algorithms tackle more complex tasks that are linked to bigger datasets, and event cameras mature to have higher spatial resolution, it is worth looking at how to encode that data efficiently when storing it on disk. To give you an example, Prophesee’s latest automotive object detection dataset is some 3.5 TB in size for under 40h of recordings with a single camera.

Spiking neurons: a digital hardware implementation

Tue, 03 Jan 2023 00:00:00 +0000

In this article, we will try to model a layer of Leaky Integrate and Fire (LIF) spiking neurons using digital hardware: registers, memories, adders and so on. To do so, we will consider a single output neuron connected to multiple input neurons from a previous layer.

In a Spiking Neural Network (SNN), neurons communicate by means of spikes: these activation voltages are then converted to currents through the synapses, charging the membrane potential of the destination neuron. In the following, the destination neuron is denoted as post-synaptic neuron, with the index $i$, while the input neuron under consideration is denoted as pre-synaptic neuron, with the index $j$.

About

Mon, 01 Jan 0001 00:00:00 +0000

I come from a pretty small village in the south of Italy. Since I was a kid, I have been interested in science and computers.

I moved to Torino for my studies. Here I did my Ph.D. in electronics engineering, under the supervision of Professor Luciano Lavagno and Professor Mario R. Casu, working on high level synthesis of digital circuits and deep learning acceleration on FPGAs.

Now I am working as GPUAI compiler engineer at Qualcomm.

Updates

Mon, 01 Jan 0001 00:00:00 +0000

2023-09-13

I am going to give a talk the Istituto Italiano di Tecnologia, Genova, Italy, for the research group of Dr. Chiara Bartolozzi! Thanks also to Giulia D’Angelo for inviting me :)

2023-08-11

I gave a talk in Innatera about our latest position paper and my Ph.D. research. It has been really fun.

The strange dutch dish I had.

2023-08-01

Jason Eshraghian and I gave a tutorial on high level synthesis for neuromorphic computing at ICONS 2023! Check out the Github repository.