A fundamental goal of our work is to read and write the neural code in order to communicate with the brain. In order to minimize crosstalk, we typically record neurons’ electrical activity (spikes, local field potential) using capacitive electrodes, and drive their state optically using microLEDs and optogenetics. This allows us to keep our read and write modalities separate, maximizing our signal-to-noise ratios and overall device performance.
To do this, we’ve developed two families of integrated circuits (“chips”) that allow us to pack thousands of channels of stim and record into tiny footprints of area and power. We’ve developed an ultra-miniaturized electrical recording frontend we call a nixel, short for neural interface element, and of course we already have a word for an optical stimulation element: a pixel.
Together, these unlock groundbreaking new opportunities in neural interface device development, which today we’re making commercially available for research and clinical applications by our customers and collaborators.
Nixel 512
Our nixels are fully differential, low power amplifiers with integrated analog-to-digital conversion. The first chip in this family is Nixel 512, which can record 512 simultaneous channels through 256 nixels.
Think of each one as a pixel, but for neural signals rather than images. A single nixel is just an element, like a dot in a picture, but many nixels can encode whole thoughts or actions. Where an image sensor reads out frames of images, an array of nixels reads out frames of brain activity.
| Input referred noise | As low as 4.5 uV |
| ADC resolution | Up to 16 bits |
| Sampling rate | Up to 32 kHz |
| Filtering | Configurable, simultaneous spike and LFP |
We think of Nixel fundamentally as a kind of electrical imager, and expect it to have uses far from our core focus, for example unlocking new possibilities in tomography or other applications that demand massively parallel, high speed, low noise electrical detection and digitization.
The nixel chip contains an array of 256 nixels, each composed of two electrodes, a low-noise amplifier, analog memory, and a comparator which can record simultaneously 512 channels fully differentially in 16 mm2, making it one of the highest density neural recording ICs available to researchers.
Full documentation for Nixel 512, including a downloadable datasheet, is available on our Docs site.
Pixel 2K and 16K
In addition to neural recording, we also need to be able to drive neuronal activity in order to convey information into the brain. To do this, we’ve developed and tested a series of ultra-miniature microLED driver chips designed specifically for optogenetics. While there are many commercially available microLED drivers out there, we’ve found that there are important differences in what makes a chip good for a smartphone or TV versus for implantation.
Today, we are releasing two Pixel optogenetic driver chips:
- Pixel 2K, an active driver with 2,048 1-bit channels in 1.8 mm2
- Pixel 16K, a passive driver with 16,384 7-bit channels in 6.5 mm2
An active driver can hold a pixel ON until told otherwise, whereas a passive driver merely scans through all of the rows and columns of pixels and activates them for the moment each is energized.
As a passive driver gets larger, the amount of time each pixel is on (assuming the full matrix is being used) decreases, which has implications for activating optogenetic proteins that often require sustained activation for some period of time. On the other hand, active drivers require more complex circuitry under each pixel — in particular a capacitor to charge and a transistor to gate it — which is significantly more complicated to fab on a thin-film probe.
Science Foundry has developed thin-film transistor processes for this purpose, as well as GaN-based microLEDs, all of which are available in our open Process Design Kits for your designs.
As with Nixel 512, full documentation for the Pixel family chips is available on our Docs site.
Get in touch
A cornerstone of our approach is to collaborate openly with the research community as well as help enable other new startups. The challenges of neural engineering are immense and we are small.
To request chips, please sign up or log into Science Foundry 
, which requires creating an organization. For this first preview run, supplies are limited and we will prioritize groups engaged in research that is complementary to our interests, defined broadly in understanding the brain and learning to engineer it. We’ll evaluate requests on a rolling basis until we run out, with the first decisions shipping at the beginning of August.
During this preview fees are $100 per chip, but no payment information is required to request allocation. If your Science Foundry organization already has linked payment information and chip allocation is granted, we will follow up and confirm your order before proceeding. Pricing for general availability will be released in the coming months.
