SQL Server 2014 :: Eliminate Used TempDB To Spill Data

I've been trying to make the following query more performant by breaking it up into smaller pieces.

SELECT MT.A3+MT.A4 AS A34,MT.A3
-- ,M.*
FROM
Master_TAB M
JOIN (SELECT M.A1,t3.A3,t3.A4,M.A6,M.A2,ROW_NUMBER() OVER (PARTITION BY A1,A6,A3,A4 ORDER BY A5 DESC) AS rownum

[Code] ....

I know that the Spill is caused by the Sort but I can't remove the sort (sort can't be done in front end). My master table had 1.7 million rows and almost 200 columns (bad design? I know but can't be changed as there's too much that would be affected) every row is little over 1KB

Here's my attempt...

-- MASTER_TAB has 1.7 million rows and 50 columns
CREATE TABLE [dbo].[tmp_ABC](
[A1] [varchar](13) NOT NULL,
[A2] [varchar](5) NOT NULL,
[A3] [varchar](4) NOT NULL,
[A4] [varchar](4) NOT NULL,
[A5] [int] NULL
) ON [PRIMARY]

[Code] ...

This is the Query that is causing the Spill (in reality I'm supposed to bring back all 200 columns fro the master table but for debug purposes I limited the columns)

Select c.A3+c.A4 as A34, c.A3, c.A1
-- M.*
from tmp_DEF c
join MASTER_TAB M on M.A1 = c.A1 and M.A2 = c.A2
order by c.A3, C.A4

if I just run the following I get no spill:

Select c.A3+c.A4 as A34, c.A3, c.A1
from tmp_DEF c
order by c.A3, C.A4

as soon as I add the Master table as a Join I get the Spill...

I read many articles, tried many suggested things (creating indexes... clustered, non-clustered) without success. Maybe I'm totally in Left Field and should enhance the performance going another route?