The Analog Paradigm

Anabrid is a technology company that essentially builds ontop of the analog paradigm. This is to make use of mathematical analogies at constructing high performing computer architectures. These architectures are currently either digital, analog or both and can incorporate novel materials, photonics or quantum properties in the future.

Resurgence of Analog Systems

Analog information processing offers remarkable advantages in computing time, latency, energy efficiency, heat dissipation, and system complexity. As digital system engineering faces fundamental limitations—such as stagnating single-core clock speeds and thermal constraints—the need for novel, unconventional computing architectures has become more pressing. Analog systems, long relegated to I/O applications, are now poised for a resurgence. Despite being overshadowed by digital computing, analog electronics never truly disappeared. Given its rich history and maturity, particularly from the golden age of analog computing, it is well-positioned to take on a more prominent role in modern computing.

Analog technology is often misunderstood, with common misconceptions labeling it as imprecise, bulky, prone to noise and faults, or even outdated—sometimes equated with being “non-digital” or “non-electronic.” In truth, these assumptions are largely unfounded. Anabrid presents compelling evidence to dispel these myths and showcase the true capabilities of analog systems.

Analog computers come in various forms, each suited to different modeling approaches. Anabrid specializes in classical operational amplifier-based electrical-analog computing, prioritizing continuous-time and continuous-value representation without relying on switched capacitor matrices. Additionally, we explore alternative analog approaches, such as purely digital Digital Differential Analyzers (DDAs)—data flow machines that run efficiently on FPGAs. While DDAs enable fully parallel digital architectures, they do so at the cost of analog computing’s inherent energy efficiency.

Anabrid Analog-Digital Hybrid Computing

Anabrids unique position in analog hardware is that we develop systems which are using purely analog information processing with digital steering. This allows to leverage key performance metrics such as low latency, high speed and low energy consumption. This is not possible with systems which have semi-analog information processing options such as switched capacitor matrices.

Our approach is unique because we have build product-level discrete products which already demonstrate key characteristics and just need to be miniaturized onto a microchip. We believe that we have to keep compatibility to well-established CMOS production processes in order to be able to reach the goal of a co-processor sitting on the same die as a digital processor. This is why we concentrate on an affordable 65nm processes which can be produced all over the place in fabs in Germany and Europe.

In our approach, the digital host system serves as the control unit for the analog computer. This process is analogous to a lab-on-a-chip (LOC), where we configure a precisely controlled electronic “experiment” and allow it to evolve to solve the given mathematical problem. When tightly integrated into CMOS, this computation will occur within just a few cycles of the digital processor. Similar to the way GPUs operate, this means the CPU remains idle while the coprocessor is actively computing.

Anabrids unique position in analog computing is that it is one of the few players that do general-purpose analog computing, including any kind of nonlinear operations which we can do purely on the analog domain. This contrasts special purpose approaches such as purely linear algebra accelerators, for instance for matrix multiplication.

Applications in Artificial Intelligence

Approaches to unconventional computing are neccessary for many challenging or intractable computational problems these days. One of the most pressing problems is training and inference of AI systems such as large language models or AI image generation. Typically, the performance of AI systems is limited by the energy and memory barriers given the large heating of digital processors and the architectural bottleneck between memory and computation. The branch of neuromorphic computing tries to overcome the inefficient way how AI computation is done these days by bionic design. On example are in-memory computer architectures which weave computational processing and information retention. In fact, in neuroscience, there is a strong belief that major processes in the human brain happen in the analog domain and not the digital domain. Many stakeholders in the AI domain believe that a general AI breakthrought will only happen with all-analog architectures. At anabrid, we present a few details on the disruptive properties of all-analog AI in the following whitepapers:

Anabrid has a number of technical approaches to accelerate AI better then anybody else, for instance:

We believe that nonlinearity and recurrence are key ingredients to build powerful AI systems with generative/forecasting powers which are beyond contemporary deep learning.

However, remarks have to be made: First, AI is currently not in our main focus. We are not a classical hardware AI startup but instead think that general purpose analog computing has a much wider set of area. In order to explain this thesis, we state that mapping any computational problem to methods from machine learning and artificial intelligence is not ideal for many reasons. First, the vast majority of AI methods is based on statistics, i.e. it accepts inaccuracy by design. However, in many areas, computationally introduces inaccuracy is not acceptable. In contrast, many computational problems have a direct mathematical equivalent which can be mapped whith high precision on analog computers which provide fast times to solution. This argument concludes the technical pitch for general purpose computing.

Analog Inspired Quantum Computing

Quantum computing these days is sometimes even used as a synonym for solving intracable problems by combining various unconventional large scale computing methods. Technically, the term refers to computers that exploit the properties of quantum mechanics in order to massively speed up some sort of computations.

Quantum computers and classical analog computers share many properties: They are both non-algorithmic but programmed by composing circuits. They both rely on careful preparation of an initial state before “letting nature” evolve that state into a set of target states which represent the solution of the problem. Both paradigms exploit the 100% full parallelity and (non-digital, unclocked) continuity of a physical system. And they are both not beyond Turing computation and they both compete to solve mankind’s greatest intracable computational problems.

Despite some people argue that a quantum computer is also an analogy computer, technologically the biggest difference between the classical analog computer by anabrid and the prototypical quantum computer is that quantum computers can exploit entanglement, thus having access to a huge, sometimes called “exponential” parameter space. However, there is a whole branch of science about quantum inspired classical computing and it might be in fact that classical annealers can simulate entanglement by coupled nonlinear systems as well as quantum computers could do.

One of the major drawbacks in quantum computing is its long time until practical use. Depending on the technology, fundamental material research is neccessary before progress can be made. When it comes to operations, many quantum computers require expensive cooling close to absolute zero. These laboratory conditions can probably only be provided in data center labs.

At anabrid, we embrace quantum computing as a technological ingredient in our roadmap. In the mean time, we build machines which can compete with contemporary quantum computers. The biggest undertaking in this direction is the REDAC product which was realized within the German national DLR Quantum Computing Call. This machine will hopefully demonstrate analog supremacy compared to digital systems and be able to simulate 10-20 qubits with perfect fidelity in 2025.