Questions? philiprbrenan@appaapps.com
Last updated: 2024-05-15 at 22:37:41 Z
Chip factories and data centers can consume 100 times more power than a typical industrial business. Wall Street Journal 2024-03-29 A14
Electricity demand to power data centers is projected to increase by 13% to 15% com- pounded annually through 2030. Yet a shortage of power is already delaying new data centers by two to six years, according to commercial- real estate firm CBRE Group. It is also driving Big Tech companies into the energy business. Amazon this month struck a $650 million deal to buy a data center in Pennsylvania powered by an on-site 2.5 gigawatt nuclear plant. Wall Street Journal 2024-03-29 A14
High speed, low power database system on a chip. The need for faster graphics produced the GPU industry. The need to secure crypto currencies quickly produced the bitcoin miner industry. Most commercial data processing is about looking stuff up. There is an opportunity to build application specific integrated circuit chips that do database look-ups quickly and to sell these chips to Cloud Computing companies solely on the basis that they will lower their electricity bills.
An application specific integrated circuit mask for a chip dedicated to high speed data look up and a working chip produced from that mask. Obviously a working Silicon sample whose performance could be measured would be a crucial proof that Database on a Chip has value.
Using semiconductors to perform look ups in nanoseconds rather then using software to do such look ups in microseconds.
Layers of flash memory and Static Random Access Memory integrated in a vertical stack with close to memory processing using conventional digital logic gates to perform high speed database look ups in parallel on the stored data one thousand times faster than can be done in software and using one thousandth of the electrical power. Chips connected to each other via optical Ethernet in parallel to provide scale with Terahertz transmission rates between them and with the outside world.
Cloud Computing data centers with large electricity bills.
Perhaps savings on the order of $100 million per year.
In rough terms, it might be possible to reduce global data center electricity usage by 1% which might lead to potential savings of $100 million per year. Such savings would be split with the data centers to give them an incentive to adopt this technology while creating profits to reward investors to drive new designs.
The initial sales would be driven by real, hard money savings resulting from reduced electricity bills. But, being able to look things up quickly, faster than a competitor, might well bestow unexpected other benefits which might make such chips valuable in as yet unknown applications, perhaps, for example: analyzing the movements of the stock market for financial companies. If such unexpected benefits did materialize then the capability to produce such chips at scale would be that much more the profitable.
There are also softer benefits like being able to claim green credentials in advertising cloud computing. Gmail is apparently run partly on hydro electricity from Bonneville Dam .
According to the IEA world production of electricity is: 10 trillion kilowatt hours per year. Data centers use about 1% of that electricity. Say we could reduce data center electricity usage by 1% that would save 1 billion kilowatt hours annually. A kilowatt hour costs approximately 10 cents. This would yield gross savings of $100 million per year.
Splitting the savings with the Cloud Computing data center industry ensures that they make savings of $50 million per year which might be enough to incentivize them to use this technology while leaving room for $50 million in profits annually for the mask rights holder.
Payments for the right to use the mask describing the application specific integrated circuit to produce chips for sales to Cloud Computing data centers.
The alternative is to continue to use conventional generic CPUs to look up data in conventional memory. While Moore's Law was still in effect it was too expensive to develop specialist techniques for data look up because the generic look up techniques provided by conventional CPUs were improving so rapidly. Now that Moore's Law has run its course there are opening for such specialist solutions: no doubt some-one will eventually take this opportunity and apply it to databases.
Both Oracle and IBM have products that indicate that they have given thought to this problem.
DB2 with BLU Acceleration runs on existing IBM infrastructure. It is a software solution not a chip level hardware solution.
Oracle Exadata is a conventional computer optimized with appropriate memory, disk drives and network options to provide an optimal conventional computing environment in which to run an Oracle database. But there is no evidence of any new chip design being used to power this solution.
In summary, both of these important database vendors are aware that hardware acceleration is beneficial to database processing but they have not moved beyond the constraints of conventional solutions. Given that both companies operate in mature markets, a break through on this area would surely have to be of great interest to them.
Other companies are also attempting to address this issue from the software side: DataEngine The DataEngine is a containerized computing environment that customers deploy on their choice of CPUs, GPUs and DPUs. Obviously they would be much better off if they could do this in hardware instead of software.
Likewise PineCone which is using in memory databases in much the same way as BLU acceleration. Using conventional memory is not enough, they should be adding logic, in situ, to the memory so processing can be done quickly in hardware close to the memory rather than slowly in conventional hardware far from the memory.
Influx data offers sub-second queries so again their implementation must be a software one. We want to offer nano second queries so our solution must be a hardware one.
My personal favorite: bit.io - full marks for a great name while using Postgres but still doing things in software not in Silicon .
All of these companies like to mention that they make artificial intelligence applications easier and faster. We could too!
The very real possibility of saving hard money in real Cloud Computing centers by reducing the time and electricity needed to perform database look-ups by a factor of one thousand.
Mask for application specific integrated circuit implementing Database on a Chip.
Send a fax to each of the major Cloud Computing data center operators - Amazon Amazon Web Services, Microsoft Azure, Google Cloud Platform etc. explaining the benefits with a sample chip mounted on a printed circuit board following in the mail for them to test by plugging the printed circuit board into an Ethernet router and doing some fast look ups.
Copies to Oracle to keep them apprised of the situation.
Plug a printed circuit board board with such a chip mounted on it into an Ethernet router and show that it quickly answers queries against a preloaded data set.
Construct proposed mask in software and test through emulation. This task has been done using our purpose built Electronic Design Automation tools but needs to be cross checked with industry standard tools to confirm expected operation.
Use the mask to create a demo chip in hardware and mount on a printed circuit board
Send the printed circuit board to well known data center operators for evaluation
Report the real savings from one customer to persuade more customers that they must follow
Sustained production at TSMC ltd in Arizona to supply follow on demand
None currently. Desired: Organization that can provide expertise in converting a design emulated in software to one realized in hardware.
We are looking for a series A round with a 10% raise of $1 million against a valuation of $10 million.
We have produced a Graphic Design System 2 file representing the mask of our proposed chip. The mask represents a digital circuit that looks up a key in a B-Tree . Graphic Design System 2 is the de facto industry standard for Electronic Design Automation data exchange of integrated circuits.
Our design has been verified by emulating the actions of the low level digital logic gates used in its construction. We need help converting this mask into a working chip that we can send to people for evaluation.
Here is our digital logic gate level design of a chip that does fast database look-ups:
Gate level tape out
That some well funded company will seize on this idea before we have made enough progress at the hardware level and get there first.
Expertise in the basics of chip manufacture to help us produce a working chip from our existing design plus all the usual stuff that goes with that: tools, personnel skilled in using these tools, the means to pay these people, financial backing, funding advice etc.
Unknown
None
Unknown.
None
Unknown.
We probably do not need office space as this work can be done anywhere in the world including, for the most part, from home offices as long as they as equipped with a usable workstation and high speed Internet access.
Oracle might want to buy this intellectual property just like they bought MySql because doing so would give them a jump start on the technology while denying that capability to their competitors in an otherwise mature industry.
Our Silicon Chip Electronic Design Automation software is written entirely in Java .