AI & Signal Processing for In-Memory Computing

At Synthara, I work in a customer-facing engineering setting, supporting pre-sales engagements by translating stakeholder requirements into KPI-driven analysis, and owning pipelines that automate workload benchmarking.

Situation

Synthara provides the world’s first commercially available, silicon-proven, in-memory computing (IMC) IP solutions. Synthara’s technology operates on the matrix-product primitive, unlocking over 100x speed and energy efficiency improvements vs. comparable hardware by reducing the need to move data between the memory and processor. IMC has existed for some time, yet practical adoption has lagged for various reasons, making commercialisation challenging.

In-Memory Computing (IMC) moves computations from the processor to the memory, eliminating the memory bottleneck and enabling faster, more efficient data processing.

Task

Working in a customer-facing engineering role, I support three areas driving Synthara’s commercial efforts:

1. Engineering an automated pipeline for benchmarking AI, signal processing, and other matrix-algebra workloads with an in-house transaction-accurate emulator, and contributing application-layer features to the emulator itself.
2. Supporting pre-sales engagements by translating customer requirements into KPI-driven analysis and design-space explorations, and delivering technical workshops, presentations, and collateral.
3. Contributing to team execution through planning, roadmapping, and mentoring.

Action

My work is split across three areas, with ~75% being focused on engineering and solutions, and 25% on execution and team support.

1. Engineering

After sourcing internal stakeholders’ requirements, the software development work represented my first priority. This involved three parts:

• Engineering an automated Python-based benchmarking pipeline around Synthara’s internal embedded emulator, using GCP, MongoDB, and Git CI.
• Populating and benchmarking workloads, from TinyML and deep learning models (ANNs, CNNs, RNNs) to LLMs and signal processing algorithms (FFT, STFT).
• Contributing to the emulator by adding application-layer features through mapping new layers to IMC matrix-algebra primitives, as well as debugging and performing code reviews.

The embedded emulation tool simulates microcontroller platforms, enabling rapid prototyping and granular performance metrics extraction.

2. Solutions

As the development work matured, I transitioned to supporting pre-sales engagements by:

• Translating stakeholder requirements into KPI-driven analysis using the benchmarking pipeline, performing design-space explorations, and generating tailored reports.
• Collaborating on technical collateral generation, including whitepapers, datasheets, manuals, blogs, presentations, and product value propositions.
• Supporting business development by attending international conferences, delivering presentations and workshops.

Delivering a workshop on In-Memory Computing for Edge AI at AMLD 2025.

3. Execution

In addition to engineering, at Synthara, I have:

• Contributed to team execution through lightweight planning and transitioning the small applications team from a Waterfall to a Lean Agile and Kanban methodology.
• Partnering with seniors to establish and drive the applications roadmap.
• Mentoring engineers and interns on technical and non-technical topics.

Result

I can summarise my contributions to Synthara as follows:

• The automated benchmarking pipeline and emulator allowed us to reduce benchmarking turnaround from weeks to hours, while the workload library grew from just matrix-vector products to 30+ workloads of increasing complexity.
• I have supported 15+ pre-sales engagements and the generation of 60+ B2B and VC leads, contributing to the signing of the first revenue-generating contracts.
• Through planning, roadmapping, and mentorship, I have helped improve the team quality and throughput, enabling smoother project execution and delivery.

Scars

Coming out of my PhD, I was used to deep work in isolation, believing that great technical solutions would sell themselves. Synthara taught me several key lessons:

• Always design for modularity, as needs shift, making monolithic solutions either obsolete or hard to maintain. Modularity favours scalability and adaptability, which, together with lightweight Agile and proactive roadmapping, help teams stay on top of features and priorities.
• Talk in terms of the audience’s needs and objectives. Taking
  an interest in your counterpart, their needs and objectives, and framing your solution in those terms is crucial to building trust and securing buy-in. Also, while training materials help, nothing beats real-world experience to build skills like active listening, empathy, and technical translation.
• Your team is your greatest asset, so invest in them and give them the tools and space to thrive. Find what drives each person’s inner greatness, and help them grow into it. Failing this easily destroys morale, and rebuilding it is hard work.

Interdisciplinary Integration

My role at Synthara covered several areas:

• AI, DSP & Data Pipeline Engineering in an embedded systems and in-memory computing context.
• Pre-sales, Technical Marketing, and Business Development.
• Team Execution, Mentorship, and Roadmapping.

Technologies & Skills Used

Technical

• Software Engineering with Python
• In-Memory Computing & Embedded Systems
• AI, LLM, Cryptography, Signal Processing Algorithm Optimisation
• System-Level Emulation & Benchmarking
• Data Engineering and Analysis
• Design Thinking

Strategic & Communication

• Business Development
• Technical Marketing and Documentation
• Research Translation
• Workshop Delivery and Pitching
• Roadmapping
• Lightweight Agile/Kanban Planning
• Mentorship

Further Information

• Synthara Website