Ordinateurs fongiques: comment les champignons alimentent les puces neuromorphiques

Un nouveau type d’ordinateur biologique

Computing was initially developed with analog technology, which differs from digital technology in that it uses more complex (and messy) signals instead of clearly distinct 1 & 0.

Source: Unison Audio

L’informatique a d’abord été développée avec la technologie analogique, qui diffère de la technologie numérique en ce qu’elle utilise des signaux plus complexes (et désordonnés) au lieu de 1 et 0 clairement distincts.

En général, un signal numérique est plus facile à analyser, reproduire et transmettre. Mais un signal analogique gère mieux la complexité d’une situation du monde réel, avec toutes ses nuances.

C’est pourquoi les scientifiques se sont tournés de nouveau vers les types d’informatique analogique pour de nouveaux développements en IA, en détection et autres applications. Cela inclut de nombreux conceptions de puces dites neuromorphiques, qui imitent la façon dont le cerveau traite les données.

Un nouveau développement visant à exploiter la capacité semblable au cerveau pour effectuer du calcul est l’apparition d’ordinateurs biologiques réels, utilisant des tissus organiques pour accomplir des tâches normalement réservées aux puces en silicium. Un exemple est les organoïdes, tissus cultivés en laboratoire à partir de neurones humains, capables d’exécuter des tâches de calcul. Associé à de nouvelles techniques d’impression 3D de tissus cérébraux fonctionnels, cela pourrait ouvrir la voie à un tout nouveau type de capacités informatiques étranges.

Un autre type de composant électronique tirant parti de composants biologiques devrait être ajouté à la liste, les scientifiques de l’Ohio State University ayant créé des memristors organiques neuromorphiques, un type de processeur de données capable de se souvenir des états électriques passés. Except they created it not from neurons, but from mushrooms.

Ils ont publié leur découverte dans la revue scientifique PLOS One, sous le titre “Sustainable memristors from shiitake mycelium for high-frequency bioelectronics.”

Pourquoi utiliser l’informatique neuromorphique ?

Montée des NPU

Neural Processing Units (NPUs), also called neuromorphic chips, are a type of AI hardware that present a few advantages compared to more traditional chips like CPUs and GPUs:

• Conception plus flexible, permettant à l’architecture de la puce de s’adapter aux données d’entraînement.
• Consommation d’énergie beaucoup plus faible, parfois aussi basse que 1/100^th d’un GPU comparable.
• Une production de chaleur moindre aide à gérer le problème croissant de refroidissement qui touche les centres de données d’IA avancés.

(Vous pouvez en savoir plus sur le matériel spécialisé en IA, y compris les NPU, dans notre rapport dédié.)

“Being able to develop microchips that mimic actual neural activity means you don’t need a lot of power for standby or when the machine isn’t being used.

That’s something that can be a huge potential computational and economic advantage.”

John LaRocco – chercheur en psychiatrie à la College of Medicine de l’Ohio State.

Many methods are currently being explored for creating neuromorphic chips:

• Exploiter la ferroélectricité incipiente, un phénomène encore mal compris.
• Substrat actif utilisant du vanadium ou du titane.
• Utiliser des memristors, un nouveau type de composant électronique, qui peut exécuter des tâches d’IA à 1/800^th de la consommation d’énergie normale.

Comment les memristors imitent les synapses

Lemristors sont des composants électroniques qui imitent les synapses reliant les neurones by remembering which electric state they were toggled to after their power is turned off.

This can greatly reduce the energy and time lost from shuttling data back and forth between processors and memory.

One of the key strengths of memristors is their capacity for efficient and self-adaptive in situ learning, which is critical for applications in robotics and autonomous vehicles.

Moreover, the low power consumption of memristors is particularly beneficial in robotics and autonomous vehicles, where energy efficiency is paramount. Hybrid analog–digital memristor systems can minimize power usage during processing without sacrificing responsiveness.

The problem so far is that creating electronic memristors has relied on emerging technologies with low production yields and unreliable electronic performance, due to how recent this technology is.

Using actual neurons, like with organoids, is also an option, but neurons are actually very difficult cells to work with, being relatively fragile and hard to grow.

But neurons are not the only biological tissues capable of processing and responding to electrical signals.

One potential alternative is mycelium, the tissue constituting ordinary mushrooms, a type of organism known for its remarkable sturdiness. They can be grown with simpler bioreactors and nutrient cultures than those required for conventional neurons and neural organoids.

Construire des ordinateurs à champignons ?

Fungal materials display conductive pathways that can form dynamically under the influence of electrical stimuli, similar to the conductive filaments formed in conventional memristors.

This adaptability can lead to enhanced performance in neuromorphic applications through the facilitation of variable resistance states that mimic synaptic behaviors more closely than traditional memristive materials.

Organic materials also have the advantage of operating effectively at lower voltages while maintaining the stable switching characteristics important for memristors, even lower than for electronics memristors, themselves much less energy-consuming than traditional computing components.

This could be important for energy-efficient devices for portable electronics and Internet of Things applications that might rely on a very low energy supply.
Swipe to scroll →

Pourquoi les champignons comestibles fonctionnent pour l’informatique

The researchers used common button mushrooms, as well as edible and medicinal Shiitake mushrooms for their experiments, both species whose cultivation is well understood and cheap.

Shiitake mushrooms have previously been shown to possess a porous carbon structure when activated. This porous structure can enhance the electrochemical performance of devices, making them suitable candidates for use in energy storage systems, including supercapacitors and, potentially, memristors.

They are also very radiation-resistant, which could help for applications like aerospace, where electronic chips can be damaged by ionizing radiation like UV and solar winds.

Réponse électrique fongique

The scientists connected the test fungal mass after it was dehydrated.

Source: PLOS One

They were then tested across a range of voltages, waveforms, and frequencies for their potential memristor capabilities.

The responding analog signal was displaying strong memristive characteristics, mimicking in analog the digital signal.

Source: PLOS One

Overall, the observed rapid switching speed of 5,850 Hz, an accuracy of 90% (± 1%), relatively low energy consumption, light weight, and radiation resistance all appear to make fungal memristors attractive for edge computing, aerospace, and embedded firmware applications.

However, accuracy decreased as the frequency increased, so not all types of signals could likely be processed/computed with that method.

It should also be noted that the method creates only biodegradable materials (food-grade shiitake is grown on wood chips) and requires no rare earth or toxic materials, contrary to conventional electronic chips.

Potentiel futur

The study here was a first trial, and was limited in two ways:

• The tests were relatively short, running over only 2 months. So the long-term capacity of the fungal memristors still needs to be investigated.
• The method used a bulk production, while actual applications would need a microculture of the mycelium grown in a dedicated environment, providing much smaller and much more controlled results.

So this is really just a proof of concept, a demonstration that something as exotic as fungal computing is even possible and reliable.

Any future design would likely see the use of more consistent cultivation techniques using 3D-printed templates and structures that shape the shiitake mushroom into the desired geometry.

Programming could also be facilitated by adding electrical contacts to a 3D-printed cultivation structure.

Finally, long-term use would necessitate preservation, which could involve a variety of techniques, including dehydration, desiccation, freeze-drying, certain hydrogels, and special coatings.

Still, the idea of developing memristors with exclusively organic materials, and resistant, cheap, and biodegradable mushroom material at that, is intriguing.

Investir dans le bioprinting

BICO Group AB (BICO.ST)

As organic-based computing progresses, 3D printing of living tissues will likely become an increasingly used tool. First in research, and then for the actual production of devices leveraging this technology.

One leader in the field has been Cellink, whose machines are used for bioprinting by researchers all over the world.

Source: Cellink

In 2021, Cellink was renamed as the BICO Group, following its acquisition of Cytena in 2019 and Scienion in 2020.

Cellink is still the brand name for the bioprinting part of the business. This could also be used for creating on-demand 3D tissues or organs. (You can read a discussion on this topic in “3D Printing Human Organs – How Realistic Is It?”).

Bioprinting represents around 1/5^th of the business, with the bioscience automation segment, including imaging of biological samples, making more than 3/5^th of revenues.

Source: BICO Group AB

In the long run, bioprinting companies are likely to evolve from providing tools to researchers to becoming suppliers of pharmaceutical companies’ bioprinting therapies for patients.

This will, in turn, completely change the number of bioprinters in use and, more importantly, the volume of consumables sold every month.

This is the same process that occurred for other biolab equipment manufacturers, including genome sequencing machines from PacBio (PACB) and Illumina (ILMN), which end up making 80% of their revenues from recurring sales of consumables.

As the BICO Group is not solely dependent on this field, it can keep improving the technology until it reaches a critical mass of users, while also making money and building its sales network with bioresearchers from its other, more mature products in bioscience automation.