HPC-scale cyber-threat detection using neuromorphic processing
Pennsylvania State University
SILICON VALLEY R&D
|
Fourteen federal programs
Applied AI research for high-stakes defense.
Twelve years building what the Department of Defense and the Department of Energy ask for and cannot buy. Real-time cybersecurity, neuromorphic computing, hyperspectral imaging and radar, taken from first principles through to fielded hardware and software.
Navy, DoE and DARPA
Neuromorphic AI research
Cleared federal vendor
U.S. patented innovations
Core pillars
Silicon Valley research, built for work that has to hold.
A second layer, not a replacement
Our cybersecurity work runs alongside the perimeter tools an organisation already has. It is built to reduce the likelihood that an intrusion succeeds, and to surface the ones that do sooner.
Hybrid deep learningNeuromorphic computeGPU acceleration
Research that ships
Fourteen federal programs across cybersecurity, imaging, radar, materials and hypersonics. Each one starts as a problem with no answer yet and ends as something that works outside a laboratory.
Backed and partnered
Work delivered with Lockheed Martin, Idaho National Laboratory, Penn State, Duke and North Carolina State, under Navy, Air Force, DARPA, MDA, DHS and Department of Energy sponsorship.
End to end, in one building
We design the method, build the hardware and software around it, and carry it through to something a program office can actually field. No handoff to a third party in the middle.
Hyperspectral AIMetamaterialsMobile hardeningMetal recovery
Deployed software and systems
One product line, end to end.
Real-time network defense through to source code, plus the materials and mobile-security work that came out of the same research floor.
Flagship · GPU and neuromorphic ready
CyberNeuro-RT
Real-time network defense that learns what normal traffic looks like on your network, then flags what does not fit. The same models run in a cloud core or on a low-power chip at the edge, so detection happens where the traffic is.
Detection in real time
Traffic is scored as it arrives rather than in a batch after the fact.
Runs where you need it
One engine, deployable on CPU, GPU or neuromorphic silicon at the edge.
Learns as threats change
Models retrain continuously instead of waiting on a signature update.
Packet rate
1.42 M/s
GPU accelerated
Threat score
0.002
Within normal band
Edge draw
4.2 W
Neuromorphic node
// console output
13:42:01.002SNN core 4 · scanning TCP streamCLEARED
13:42:01.018Anomaly signature matched, session isolatedFLAGGED
13:42:01.030Rule pushed to perimeter firewallAPPLIED
Interface preview. Layout is representative of the product console; values shown are illustrative, not measured performance.
Enterprise · Autonomous response
CNRT Vanguard
CyberNeuro-RT scaled up to an operations centre. Vanguard watches an entire estate rather than a single segment, and acts on what it finds without waiting for someone to read an alert queue.
Estate-wide visibility
One picture across datacentre, campus and remote sites.
Response without the wait
Containment steps run on detection instead of after triage.
Segments
148
Monitored
Open cases
3
Under review
Auto-contained
17
Last 24 hours
Datacenter coreNominal · 94%
Campus edgeElevated · 61%
Remote sitesNominal · 88%
Interface preview. Layout is representative of the product console; values shown are illustrative, not measured performance.
DevSecOps · Source analysis
CNRT CodeLens
Finds exploitable defects in source code before the code ships. CodeLens reads a repository the way a reviewer would, ranks what it finds by how reachable it actually is, and puts the short list in front of the team that has to fix it.
Ranked by real risk
Reachable defects first, so the queue is short enough to work through.
Fits the existing pipeline
Runs against the repository on the schedule the team already keeps.
Files read
12,480
This pass
Findings
9
Ranked by reach
Reachable
2
Escalated
// console output
auth/session.c:214Unchecked bounds on copy pathREACHABLE
net/parse.c:88Integer narrowing before allocationREACHABLE
util/log.c:33Format string from callerLOW REACH
Interface preview. Layout is representative of the product console; values shown are illustrative, not measured performance.
Clean technology · Commercialised
E-RECOV
Recovers gold, copper and other metals of value from electronic waste electrochemically, at room temperature. Developed with Idaho National Laboratory and now commercialised.
More metal, less chemistry
Higher recovery than conventional processing, with far less reagent.
No toxic emissions
Runs at room temperature rather than through a smelter.
Cell temp
22 °C
Ambient
Batch
B-114
In progress
Stage
3 of 4
Elution
GoldRecovered · 96%
CopperRecovered · 91%
Reagent usedAgainst baseline · 25%
Interface preview. Layout is representative of the product console; values shown are illustrative, not measured performance.
Mobile · Kernel-level
Diamond-Droid
Hardening for Android devices at the kernel, for fleets that leave the building and still have to be trusted when they come back.
Below the app layer
Protection sits in the kernel, not in something a user can uninstall.
Built for managed fleets
Designed around devices an organisation issues and has to account for.
Devices
1,204
Enrolled
Kernel current
1,198
Patched
Quarantined
6
Held
Integrity attestedFleet · 99%
Policy currentFleet · 96%
Awaiting updateFleet · 4%
Interface preview. Layout is representative of the product console; values shown are illustrative, not measured performance.
Our latest JV
Vulcan Ventura
A joint venture, not a Quantum Ventura product and not wholly owned by us. Vulcan Ventura carries the radio-frequency and radar portfolio into programs that neither partner would take on alone.
About the joint ventureA joint ventureHeld with a partner company. Quantum Ventura is one of two owners, not the sole one.
Radio frequency and radarAntenna, array and radar work, separate from the cybersecurity product line.
Its own entityVulcan Ventura contracts in its own name and maintains its own web presence.
How the work fits together
Sense, decide, act.
The same three-step shape underneath the cybersecurity, imaging and radar work. What changes is the sensor at the front and the action at the end.
01
Sense
Network packets, hyperspectral frames, radar returns and optical sensors, taken in as they arrive.
Live input
02
Decide
Learned models score what is ordinary and what is not, running on GPUs or on low-power neuromorphic chips.
Sub-second call
03
Act
The decision reaches something that can use it, a firewall rule, an operator console, a tracking handoff.
Outcome
Federal record
Fourteen federal programs, twelve years.
Sponsored research with the Navy, the Air Force, DARPA, the Missile Defense Agency, Homeland Security and the Department of Energy, run alongside Lockheed Martin and the national laboratories.
Source-code vulnerability detection
Lockheed Martin · U. Wisconsin–Madison
Electrochemical recycling of electronic constituents of value
Idaho National Lab · Colt Refining
Real-time synthetic aperture radar simulator
Lockheed Martin
Ground-based hypersonic threat detection using neuromorphic processing
Lockheed Martin · University of Florida
AI/ML verification and validation
North Carolina State University
Hyperspectral image detection of UAVs
Lockheed Martin · Bodkin Design
Contraband detection using hyperspectral imaging and neuromorphic processing
Lockheed Martin · Bodkin Design
Metamaterial-enhanced micromirror surfaces for infrared beam control
Duke University
Smart image-recognition sensor using metamaterial optics
Lockheed Martin
Machine-learning discovery of high-entropy alloys
Multi-sensor object detection and tracking
Next-generation toolsets for weapons separation evaluation
Machine-learning database for low-flammability polymer matrix composites
No programs match that filter.
14 federal programs · 8 more behind the button
What we do
An AI research company, across seven fields.
Federal program offices bring us problems that have no answer yet. In government work you cannot be a single-focused company.
Cybersecurity
Real-time network defense and source-code analysis
Machine Learning
Neuromorphic and edge inference, verification and validation
Hyperspectral Imaging
UAV and concealed-contraband detection
Synthetic Aperture Radar
Radar and signal processing, real-time simulation
Metamaterials and MetaOptics
Infrared beam control and optical recognition
Computer Vision
Multi-sensor object detection and tracking
Hypersonics
Ground-based threat detection on neuromorphic hardware
Where we are
Five offices, three continents.
A multifaceted operation. In government work you cannot be a single-focused company.
Headquarters
San Jose, California
Quantum Ventura Inc.
Nevada
Carson City, Nevada
Operations
Middle East
Dubai, U.A.E.
Quantum Guru · Unit R40, Burjuman
Japan
Tokyo, Japan
Distribution and support
Canada
Vancouver, British Columbia
Engineering
Consortiums and certifications
Vetted, audited and on contract.
Start a conversation
Tell us the program.
Whether you are a program office with a problem, a prime looking for a subcontractor, or an enterprise evaluating CyberNeuro-RT, the fastest route is to describe what you are trying to solve.
Program officesBring the problem, not a specification. Most of our work starts before anyone knows what the answer looks like.
Primes and integratorsWe hold the consortium memberships and the audit history that subcontracting needs.
EnterprisesProduct evaluations for CyberNeuro-RT, Vanguard and CodeLens start with a briefing and a scoping call.