Industry Article

Tracking Every Neuron: The Future of Medical Research Lies in Processing Time

April 19, 2018 by Justin Kirkham

How is Stanford making use of high-speed memory in their latest research initiative?

Memory powers medical research in many ways, especially in relation to brain-mapping hardware. How is Stanford making use of high-speed memory in their latest research initiative?

When it comes to medicine, fast memory enables researchers to store and parse large datasets. One example of an exceptionally large dataset is Stanford’s MultiMAP project, which tracks, records, and analyzes the activity of nerve cells in zebrafish brains.

What Does Stanford’s MultiMAP Do?

In a Stanford press release, science writer Bruce Goldman explains that MultiMAP, Multiplexed-alignment of Molecular and Activity Phenotypes, allows researchers to “track nearly every neuron in the zebrafish brain and then identify the cell type of every neuron of interest.” The Stanford team behind MultiMAP used the technology to determine that brain circuits are, in fact, tied to alertness.


Stanford tracked the neurons in zebrafish brains, tying brain activity to alertness.

Stanford tracked the neurons in zebrafish brains, tying brain activity to alertness. Image courtesy of NICHD.


“We looked at every neuron in the fish’s brain during life, when those cells were actively firing, and learned which cells were most active at moments when we knew that the fish was most alert,” Karl Deisseroth, a Stanford bioengineering and psychiatry and behavioral sciences professor, as well as a Howard Hughes Medical Institute investigator, told Goldman. “Then, after the fish’s brain tissue was preserved with a fixative without altering relative positions of cells within the fish’s head, we could target those neurons with molecular probes and determine their cell types.” 

This technology alone functions on an enormous scale and would need a vast amount of low-latency, high-speed memory located near the processing unit in order to function. And as this sort of brain-mapping technology continues to grow and progress, that memory will have to follow suit.

MultiMAP is particularly important because of the effect the data it gathers could have on mental health. Goldman specifically notes that sleep deprivation and depression are tied to a lack of alertness, while over-alertness is tied to anxiety, mania, and PTSD.

“The new findings open the door to a whole new route of further exploration,” Deisseroth said. “The more we understand the landscape of neurons that underlie a brain state like alertness, the more we understand the brain state concept itself — and we may even be able to help design brain state targeted clinical interventions.”

In order to embark upon this “route of further exploration,” fast memory is essential. Larval zebrafish have approximately 100,000 neurons, and each of those neurons produces data that must be recorded, stored, and then quickly combed through for important data points. By the time the zebrafish reaches adulthood, it has approximately 10 million neurons, exponentially increasing the demand on Stanford’s MultiMAP technology. By comparison, a human brain contains 86 billion neurons.

Neurons in different brains chart shows that humans have over 860 times as many neurons as zebrafish.

Humans have over 860 times as many neurons as zebrafish.


What Kind of Memory Can Map a Brain?

When dealing with so much information, mapping out the brain, and identifying its decisions and functions, according to Jason Echols, Micron’s senior manager in technical marketing, researchers are entering the realm of fast data. In order to process so much data and analyze the important points therein, MultiMAP is “sort of zooming in on this AI, machine learning bubble.”

“They need to be able to do a lot of parallel processing,” Echols said. “And when they have to involve data in that decision stream, they're not just thinking about things; they're actually pulling data and analyzing it.”

Echols said medical researchers like those working on the MultiMAP project strive to “optimize their storage infrastructure to get the most performance,” especially when they’re working with the volume of data that Stanford’s researchers were looking at in brain cells.

“And that's where you're seeing people like Nvidia and others really working with them,” Echols said. “How do they get the storage to where it's cost-effective, but also super fast so that all the cores of computing that it takes to run all this stuff is fed with the data it needs? That's the true conversation.”

Pushing All Memory Forward

According to Echols, memory research powers medical research. And as streamlined, fast, and efficient memory solutions are implemented in new applications, all facets of the field will benefit.

Echols explained that of the many ways researchers and forward thinkers are implementing memory and storage on a large scale, there are eight applications that are driving the edge of science and medicine:

  • Big data and analytics
  • Internet of things (IoT) devices
  • Hybrid cloud
  • Non-volatile memory express (NVMe) and single-root input/output virtualization (SRIOV)
  • Scale-out architecture
  • Artifical intelligence (AI) and machine learning
  • Security
  • Software-defined storage (SDS)

“Just within the storage side of the business,” Echols said, “these are the ones that we're really influencing at the macro trend level.”

Per Tony Ansley, a principal technical marketing engineer at Micron, many of the applications above can benefit from the same general concept of streamlining memory: getting the data to the storage and the compute faster. When researchers can define, say, whether a neuron is active or not faster, this allows them to draw their conclusions faster, too.

“And it's not just the storage technology, but it's also the architectural technology from the entire computer industry,” Ansley said. “And how do we optimize storage for new server platforms, new server technologies, and new processor technologies that are coming out that allow us to potentially provide more storage and faster storage in a smaller overall space—to allow us to get quicker times to analytics?”


Micron Crucial DDR4 DRAM

Micron memory, like their Crucial DDR4 DRAM, helps steer medical research technology toward faster processing times. Image courtesy of Dsimic.


When data is analyzed directly on the compute, it takes about 5 nanoseconds to fully process. DRAM is extremely close to the compute in terms of processing time, taking about 30 nanoseconds. But as memory solutions appear further and further away from the compute, their processing times follow suit. A SATA SSD would take 300 microseconds to process the same data, while a SAS HDD would take 6 milliseconds. And on the far end of the spectrum, a hybrid SAN would take a full 30 milliseconds.

This may not seem like a large difference, especially with fractions of seconds as small as these figures, but Echols suggested breaking it down in “human terms.” If that data directly on the compute instead took a single second to process, DRAM would take, on the same scale, 6 seconds to process. A SATA SSD would take 16 hours to process that data, a SAS HDD would take 2 weeks, and a hybrid SAN would take 2-3 months. On this scale, Echols said, “these kinds of really fast compute architectures can't wait this long.”

“The place where Micron is actually innovating is in this space right here,” Echols said. “We own innovation in this DRAM through SSD space. And this is where we're helping move the industry forward.”

The Future of Medicine

According to Ansley, one main way in which Micron contributes to streamlined medical research technology is by implementing their memory solutions in smaller form factors, allowing researchers to go mobile.

“Especially in third world countries where there might be outbreaks of specific types of pathogens, where they're having to potentially collect real-time video scans or sonograms, it’s a lot of data, right?” Ansley said. “They don't want to have to roll a cart around. They'd much rather carry it in their hand.”


Tony Ansley believes Micron's memory is perfectly situated to power mobile medical devices.

Tony Ansley believes Micron's memory is perfectly situated to power mobile medical devices. Image courtesy of Columbus Air Force Base.


Just as MultiMAP collects data from each brain neuron, other pieces of medical technology collect incredible amounts of data, from X-ray images at high resolution to gigabytes of data in sonogram files. Researchers need more data at their fingertips to stay in the field and avoid uploading and downloading constantly. So, Ansley said, “being able to squeeze all these technologies into smaller and smaller packages more efficiently allows me to carry that in my hand instead of carrying it on my back, right?”

“Micron is part of that industry and is always looking to how do we squeeze more storage into a single NAND chip, for example,” he continued. “We were pushing what used to be a single 2.5-inch hard drive or SSD size of 100 gigs, and we're now doing 10 to 20 terabytes of data in that same form factor in just a few years. And that's going to continue.”

Industry Articles are a form of content that allows industry partners to share useful news, messages, and technology with All About Circuits readers in a way editorial content is not well suited to. All Industry Articles are subject to strict editorial guidelines with the intention of offering readers useful news, technical expertise, or stories. The viewpoints and opinions expressed in Industry Articles are those of the partner and not necessarily those of All About Circuits or its writers.

  • devid1 January 22, 2019

    I decided to try OptiMind, but not hoping for a result. On the third day I felt that I wake up myself in the morning (not according to the alarm clock), I feel good, not broken. I fall asleep well and sleep soundly, the headaches have stopped. Concentration has improved and memory.

    Like. Reply
  • Suri Montanez October 25, 2019

    One thing that has changed medical imaging the last years is the multimodal approach, which means integrating different techniques in one medical scanner like PET/MR or PET/CT or even PET/MR/EEG. This changed, of course, some steps in the processing because you can use one modality to correct the other for example using the high resolution, anatomical MRI data to motion correct the blurry PET data.

    Like. Reply