The Hardware is the Software!
The available hardware determines what and how intelligence can be engineered.
George Gilder, an American investor, powerfully described the computer chip as inscribing worlds on grains of sand. The performance of an algorithm is fundamentally intertwined with the hardware and software it runs on.
dl.acm.org/doi/epdf/10.1145/…
I had to get away from circuitry. I needed something that could arrange and rearrange at a molecular level, and to keep its form if required. Holding for memory, shifting for thoughts.
This is your hardware?
Wetware
Alex Garland, Ex Machina
link.springer.com/content/pd…
Human brains and bodies are not hardware running software: the hardware is the software. We reason that because the physics of artificial intelligence hardware and of human biological “hardware” is distinct, neuromorphic engineers need to be selective in the inspiration we take from neuroscience.
The physics of the substrates from which we construct AI systems are fundamentally different from the physics of our brains. Although often invisible, this physics (in the form of hardware strengths and constraints) has arguably been the predominant driving force for modern AI. Neuroscience inspirations have been less important (so far).
The translation of biological information processing into mathematics is an important endeavor—one that has been absolutely foundational in both theoretical neuroscience and modern AI. But this translation is not the same translation that must be performed for neuromorphic hardware because hardware is not mathematics. In our view, the forced adherence to specific mathematical models has limited the impact of neuromorphic computing.
cell.com/neuron/fulltext/S08…
Neuromorphic engineering aims to create computing hardware that mimics biological nervous systems, and it is expected to play a key role in the next era of hardware development.
nature.com/articles/s41928-0…
By controlling the amplitude/intensity, duration, and/or number of optical pulses, tunable optoelectronic memory functions, such as short-term and long-term plasticity, are experimentally established in mesoscopic α-phase vanadium pentoxide(V2O5)-based optoelectronic synapses. Device fabrication was extended to mechanically flexible ultrathin glass substrates that could be the base for applications in neuromorphic vision, such as robotics, edge electronics, distributed sensing, bioengineering, and more.
Because of the way such crystals emulate synapses, they offer a simplified circuitry that reduces both energy consumption and signal interference. They also do things our eyes cannot, like see infrared light.
advanced.onlinelibrary.wiley…
The NLR-led Research team found oxygen vacancies within the V2O5 crystals trapped charges created from incoming light, forming a so-called “polaron,” which endows the crystal with a sort of memory. As long as the charge persists, the crystal keeps a record of the light, which can then be read out with electrodes.
nlr.gov/news/detail/program/…
ISSCC 2019: Deep learning hardware: Past, present, and future - Yann LeCun.
youtube.com/watch?v=YzD7Z2yR…
Hardware limitations influence research through action: Computer scientists like to think that they think in abstract and hope that the hardware will one day support their idea, but their thinking is always limited by the hardware we have at our disposal. (11-12:30)
Yann LeCun, in this speech at the International Solid-State Circuits Conference of 2019, lays bare an often-unexplored link not only between software and hardware, but between abstraction and material implementation that traverses debates on computation.
The idea that computer scientists ‘think through hardware’ (see Munn, 2022a),
journals.sagepub.com/doi/10.… as the quote seems to suggest, was not lost on Alan Turing, for whom thinking and calculating is always performed through machines and hardware.
pmc.ncbi.nlm.nih.gov/article…